evaluating the lifetime performance index based on the bayesian
TRANSCRIPT
Hindawi Publishing CorporationJournal of Applied MathematicsVolume 2013 Article ID 547209 13 pageshttpdxdoiorg1011552013547209
Research ArticleEvaluating the Lifetime Performance Index Based onthe Bayesian Estimation for the Rayleigh Lifetime Products withthe Upper Record Values
Wen-Chuan Lee1 Jong-Wuu Wu2 Ching-Wen Hong3 and Shie-Fan Hong2
1 Department of International Business Chang Jung Christian University Tainan 71101 Taiwan2Department of Applied Mathematics National Chiayi University 300 Syuefu Road Chiayi City 60004 Taiwan3Department of Information Management Shih Chien University Kaohsiung Campus Kaohsiung 84550 Taiwan
Correspondence should be addressed to Jong-WuuWu jwwumailncyuedutw
Received 20 October 2012 Accepted 17 December 2012
Academic Editor Chong Lin
Copyright copy 2013 Wen-Chuan Lee et alThis is an open access article distributed under theCreativeCommonsAttribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited
Quality management is very important for many manufacturing industries Process capability analysis has been widely applied inthe field of quality control to monitor the performance of industrial processes Hence the lifetime performance index119862
119871is utilized
to measure the performance of product where 119871 is the lower specification limit This study constructs a Bayesian estimator of 119862119871
under a Rayleigh distribution with the upper record values The Bayesian estimations are based on squared-error loss functionlinear exponential loss function and general entropy loss function respectively Further the Bayesian estimators of 119862
119871are utilized
to construct the testing procedure for 119862119871based on a credible interval in the condition of known 119871 The proposed testing procedure
not only can handle nonnormal lifetime data but also can handle the upper record values Moreover the managers can employthe testing procedure to determine whether the lifetime performance of the Rayleigh products adheres to the required level Thehypothesis testing procedure is a quality performance assessment system in enterprise resource planning (ERP)
1 Introduction
Process capability analysis is an effective means to measurethe performance and potential capabilities of a process Pro-cess capability indices (PCIs) are utilized to assess whetherproduct quality meets the required level in manufacturingindustries For instance the lifetime of electronic compo-nents exhibits a larger-the-better type of quality character-istic Montgomery [1] proposed the process capability index119862119871to evaluate the lifetime performance of electronic com-
ponents where 119871 is the lower specification limit Tong et al[2] constructed the uniformly minimum variance unbiasedestimator (UMVUE) of119862
119871and proposed a hypothesis testing
procedure for the complete sample from a one-parameterexponential distribution In addition the lifetime perfor-mance index 119862
119871also is applied to evaluate the lifetime of
product in the censored sample For instance Hong et al [34] constructed the lifetime performance index 119862
119871to evaluate
business performance under the right type II censored sample
and proposed a confidence interval for Paretorsquos distributionWu et al [5] proposed a hypothesis testing procedure basedon a maximum likelihood estimator (MLE) of 119862
119871to evaluate
the product quality for two-parameter exponential distri-bution under the right type II censored sample Lee et al[6] also proposed a hypothesis testing procedure based ona MLE of 119862
119871to evaluate product quality for exponential
distribution under the progressively type II right censoredsample Lee et al [7] also constructed an UMVUE of 119862
119871
and developed a testing procedure for the performance indexof products with the exponential distribution based on thetype II right censored sample All of the above 119862
119871have been
utilized to evaluate the quality performance for complete dataand censored data Nevertheless record values often arise inindustrial stress testing and other similar situations
Record values and the associated statistics are of interestand important in many real-life applications In industry andreliability studies many products fail under stress For exam-ple a battery dies under the stress of time But the precise
2 Journal of Applied Mathematics
breaking stress or failure point varies even among identicalitems Hence in such experiments measurements may bemade sequentially and only the record values (lower or upper)are observed Record values arise naturally in many real-life applications involving data relating to weather sportseconomics and life tests According to themodel of Chandler[8] there are some situations in lifetime testing experimentswhere the failure time of a product is recorded if it exceedsall preceding failure times These recorded failure times arethe upper record value sequence In general let 119883
1 1198832
be a sequence of independent and identically distributed(iid) random variables having the same distribution asthe (population) random variable 119883 with the cumulativedistribution function (cdf) 119865(119909) and the probability densityfunction (pdf) 119891(119909) An observation 119883
119895will be called an
upper record value if it exceeds in value all of the precedingobservations that is if119883
119895gt 119883119894 for every 119894 lt 119895 The sequence
of record times 119879119899 119899 ge 1 is defined as follows
Let 1198791= 1 with probability 1 for 119899 ge 2 119879
119899= min119895
119883119895gt 119883119879119899minus1
A sequence of upper record values 119883119880(119899) is
then defined by
119883119880(119899)
= 119883119879119899
119899 = 1 2 (1)
Let 119883sim= 119883119880(1) 119883119880(2) 119883
119880(119899) be the first 119899 upper record
values arising from a sequence of iid random variables1198831 1198832 with cdf 119865(119909)
In this study we consider the case of the upper recordvalues instead of complete data or censored data In orderto evaluate the lifetime performance of nonnormal datawith upper record values the lifetime performance index 119862
119871
is utilized to measure product quality under the Rayleighdistribution with the upper record values The Rayleighdistribution is a nonnormal distribution and a special case ofthe Weibull distribution which provides a population modeluseful in several areas of statistics including life testing andreliability whose age with time as its failure rate is a linearfunction of time Bhattacharya and Tyagi [9] mentioned thatin some clinical studies dealing with cancer patients thesurvival pattern follows the Rayleigh distribution Cliff andOrd [10] showed that the Rayleigh distribution arises as thedistribution of the distance between an individual and itsnearest neighbor when the special pattern is generated by thePoisson process Dyer and Whisenand [11] demonstrated theimportance of this distribution in communication engineer-ing (also see Soliman and Al-Aboud [12])The pdf and cdfof the Rayleigh distribution are given respectively by
119891 (119909) =119909
1205792119890minus1199092
21205792
(2)
119865 (119909) = 1 minus 119890minus1199092
21205792
(3)
where 119909 gt 0 and 120579 gt 0 respectively When119883 comes from theRayleigh distribution the mean of 119883 is 119864(119883) = radic1205872120579 andthe standard deviation of119883 is 120590 = radic(4 minus 120587)2120579
Bayesian and non-Bayesian approaches have been used toobtain the estimators of the parameter 120579 under the Rayleigh
distribution with the upper record values Soliman and Al-Aboud [12] compared the performance of the Bayesian esti-mators with non-Bayesian estimators such as the MLE andthe best linear unbiased (BLUE) estimatorThe Bayesian esti-mators are developed under symmetric and nonsymmetricloss functions The symmetric loss function is squared-error(SE) loss function The nonsymmetric loss function includeslinear exponential (LINEX) and general entropy (GE) lossfunctions In recent decades the Bayesian viewpoint hasreceived frequent attention for analyzing failure data andother time-to-event data and has been often proposed as avalid alternative to traditional statistical perspectives TheBayesian approach to estimation of the parameters andreliability analysis allows prior subjective knowledge onlifetime parameters and technical information on the failuremechanism as well as experimental data Bayesian methodsusually require less sample data to achieve the same qualityof inferences than methods based on sampling theory (alsosee [12])
The main aim of this study will construct the Bayesianestimator of 119862
119871under a Rayleigh distribution with upper
record values The Bayesian estimators of 119862119871are developed
under symmetric and nonsymmetric loss functions Theestimators of 119862
119871are then utilized to develop a credible
interval respectively The new testing procedures of credibleinterval for 119862
119871can be employed by managers to assess
whether the products performance adheres to the requiredlevel in the condition of known 119871 The new proposed testingprocedures can handle nonnormal lifetime data with upperrecord values Moreover we will evaluate the performance ofthe new proposed testing procedures with Bayesian and non-Bayesian approaches
The rest of this study is organized as follows Section 2introduces some properties of the lifetime performanceindex for lifetime of product with the Rayleigh distributionSection 3 discusses the relationship between the lifetimeperformance index and conforming rate Section 4 developsa hypothesis testing procedure for the lifetime performanceindex119862
119871with the non-Bayesian approach Section 5 develops
a hypothesis testing procedure for the lifetime performanceindex 119862
119871with the Bayesian approach Section 6 discusses the
MonteCarlo simulation algorithmof confidence (or credible)level One numerical example and concluding remarks aremade in Sections 7 and 8 respectively
2 The Lifetime Performance Index
Montgomery [1] has developed a process capability index 119862119871
to measure the larger-the-better quality characteristic Then119862119871is defined by
119862119871=120583 minus 119871
120590 (4)
where 120583 120590 and 119871 are the process mean the process standarddeviation and the lower specification limit respectively
To assess the product performance of products 119862119871can
be defined as the lifetime performance index If 119883 comesfrom theRayleigh distribution then the lifetime performance
Journal of Applied Mathematics 3
index 119862119871can be rewritten as
119862119871=120583 minus 119871
120590=radic1205872120579 minus 119871
radic(4 minus 120587) 2120579= radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579
minusinfin lt 119862119871lt radic
120587
4 minus 120587
(5)
where 120583 = 119864(119883) = radic1205872120579 is the process mean 120590 =
radicVar(119883) = radic(4 minus 120587)2120579 is the process standard deviationand 119871 is the lower specification limit
3 The Conforming Rate
If the lifetime of a product 119883 exceeds the lower specificationlimit 119871 then the product is defined as a conforming productThe ratio of conforming products is known as the conformingrate and can be defined as
119875119903= 119875 (119883 ge 119871) = 119890
minus(12)(radic1205872minusradic(4minus120587)2119862119871)2
minus infin lt 119862119871lt radic120587 (4 minus 120587)
(6)
Obviously a strictly increasing relationship existsbetween the conforming rate 119875
119903and the lifetime
performance index 119862119871 Table 1 lists various 119862
119871values and
the corresponding conforming rate 119875119903by using STATISTICA
software [13] (also see [14])For the 119862
119871values which are not listed in Table 1 the
conforming rate 119875119903can be obtained through interpolation In
addition since a one-to-one mathematical relationship existsbetween the conforming rate119875
119903and the lifetime performance
index 119862119871 Therefore utilizing the one-to-one relationship
between 119875119903and 119862
119871 lifetime performance index can be
a flexible and effective tool not only evaluating productperformance but also estimating the conforming rate 119875
119903
4 Testing Procedure for the LifetimePerformance Index 119862
119871with
Non-Bayesian Approach
This section will apply non-Bayesian approach to construct amaximum likelihood estimator (MLE) of119862
119871under aRayleigh
distribution with upper record values The MLE of 119862119871is
then utilized to develop a new hypothesis testing procedurein the condition of known 119871 Assuming that the requiredindex value of lifetime performance is larger than 119888
0 where
1198880denotes the target value the null hypothesis 119867
0 119862119871le 1198880
and the alternative hypothesis1198671 119862119871gt 1198880are constructed
Based on the new hypothesis testing procedure the lifetimeperformance of products is easy to assess
Let 119883 be the lifetime of such a product and 119883 hasa Rayleigh distribution with the pdf as given by (2)Let 119883sim
= (119883119880(1) 119883119880(2) 119883
119880(119899)) be the first 119899 upper
record values arising from a sequence of iid Rayleigh
variables with pdf as given by (2) Since the joint pdf of(119883119880(1) 119883119880(2) 119883
119880(119899)) is
119891 (119909119880(119899))
119899minus1
prod
119894=1
119891 (119909119880(119894))
1 minus 119865 (119909119880(119894)) (7)
where 119891(119909) and 119865(119909) are the pdf and cdf of 119883 respec-tively (also see [12 15 16]) So the likelihood function of(119883119880(1) 119883119880(2) 119883
119880(119899)) is given as
119871 (120579 | 119909119880(1) 119909
119880(119899)) =
119899
prod
119894=1
119909119880(119894)
1205792 exp(minus
1199092
119880(119899)
21205792) (8)
The natural logarithm of the likelihood function with (8) is
ln 119871 (120579 | 119909119880(1) 119909
119880(119899)) =
119899
sum
119894=1
ln (119909119880(119894)) minus 2119899 ln 120579 minus
1199092
119880(119899)
21205792
(9)
Upon differentiating (9) with respect to 120579 and equating theresult to zero theMLE of the parameter 120579 can be shown to be
120579MLE =119883119880(119899)
radic2119899(10)
(also see [12]) By using the invariance of MLE (see Zehna[17]) the MLE of 119862
119871can be written as given by
119862119871MLE = radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579MLE
= radic120587
4 minus 120587minus radic
2
4 minus 120587sdot 119871 sdot
radic2119899
119883119880(119899)
(11)
Construct a statistical testing procedure to assess whetherthe lifetime performance index adheres to the required levelThe one-sided confidence interval for 119862
119871is obtained using
the pivotal quantity 1198832119880(119899)1205792 By using the pivotal quantity
1198832
119880(119899)1205792 given the specified significance level 120572 the level
100(1 minus 120572) one-sided confidence interval for 119862119871can be
derived as followsSince the pivotal quantity 1198832
119880(119899)1205792sim 1205942
2119899 and 1205942
21198991minus120572
function which represents the lower 1 minus 120572 percentile of 12059422119899
then
119875(1198832
119880(119899)
1205792le 1205942
21198991minus120572) = 1 minus 120572
997904rArr 119875(119862119871ge radic
120587
4 minus 120587minus(radic
120587
4 minus 120587minus119862119871MLE)(
1205942
21198991minus120572
2119899)
12
)
= 1 minus 120572
(12)
where 119862119871= radic120587(4 minus 120587) minus radic2(4 minus 120587)(119871120579) From (12) we
4 Journal of Applied Mathematics
Table 1 The lifetime performance index 119862119871versus the conforming rate 119875
119903
119862119871
119875119903
119862119871
119875119903
119862119871
119875119903
minusinfin 0000000 040 0611832 120 0896627minus450 0000147 045 0631685 125 0909965minus400 0000551 050 0651484 130 0922511minus350 0001858 055 0671182 135 0934227minus300 0005628 060 0690734 140 0945076minus250 0015308 065 0710094 145 0955027minus200 0037404 070 0729213 150 0964047minus150 0082094 075 0748044 155 0972109minus100 0161849 080 0766539 160 0979187minus050 0286621 085 0784648 165 0985259000 0455938 090 0802324 170 0990306015 0513214 095 0819518 175 0994310020 0532718 100 0836183 180 0997261025 0552370 105 0852271 185 0999147030 0572132 110 0867738 190 0999963035 0591967 115 0882538 191 0999997
obtain that a 100(1 minus 120572) one-sided confidence interval for119862119871is
119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871MLE)(
1205942
21198991minus120572
2119899)
12
(13)
where119862119871MLE is given by (11)Therefore the 100(1minus120572) lower
confidence interval bound for 119862119871can be written as
LBMLE = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871MLE)(
1205942
21198991minus120572
2119899)
12
(14)
where 119862119871MLE 120572 and 119899 denote the MLE of 119862
119871 the specified
significance level and the upper record sample of sizerespectively
The managers can use the one-sided confidence intervalto determine whether the product performance adheres tothe required levelThe proposed testing procedure of119862
119871with
119862119871MLE can be organized as follows
Step 1 Determine the lower lifetime limit 119871 for productsand the performance index value 119888
0 then the testing null
hypothesis 1198670 119862119871le 1198880and the alternative hypothesis
1198671 119862119871gt 1198880are constructed
Step 2 Specify a significance level 120572
Step 3 Given the upper record values 119909sim= (119909119880(1) 119909119880(2)
119909119880(119899)) the lower lifetime limit 119871 and the significance level 120572
then we can calculate the 100(1 minus 120572) one-sided confidenceinterval [LBMLEinfin) for 119862119871 where LBMLE as the definition of(14)
Step 4 The decision rule of statistical test is provided asfollows
(1) if the performance index value 1198880notin [LBMLEinfin) then
we will reject 1198670 It is concluded that the lifetime
performance index of products meets the requiredlevel
(2) if the performance index value 1198880isin [LBMLEinfin) then
we do not reject 1198670 It is concluded that the lifetime
performance index of products does not meet therequired level
5 Testing Procedure forthe Lifetime Performance Index 119862
119871with
the Bayesian Approach
This section will apply the Bayesian approach to constructan estimator of 119862
119871under a Rayleigh distribution with upper
record values The Bayesian estimators are developed undersymmetric and nonsymmetric loss functionsThe symmetricloss function is squared-error (SE) loss function The non-symmetric loss function includes linear exponential (LINEX)and general entropy (GE) loss functions The Bayesian esti-mator of 119862
119871is then utilized to develop a new hypothesis
testing procedure in the condition of known 119871 Assumingthat the required index value of lifetime performance is largerthan 119888
0 where 119888
0denotes the target value the null hypothesis
1198670 119862119871le 1198880and the alternative hypothesis119867
1 119862119871gt 1198880are
constructed Based on the new hypothesis testing procedurethe lifetime performance of products is easy to assess
51 Testing Procedure for the Lifetime Performance Index119862119871with the Bayesian Estimator under Squared-Error Loss
Function Let 119883 be the lifetime of such a product and 119883has a Rayleigh distribution with the pdf as given by (2)We consider the conjugate prior distribution of the formwhich is defined by the square-root inverted-gamma density
Journal of Applied Mathematics 5
as follows
1205871 (120579) =
119886119887
Γ (119887) 2119887minus1120579minus2119887minus1
119890minus1198862120579
2
(15)
where 120579 gt 0 119886 gt 0 and 119887 gt 0
119864 (120579) = (119886
2)
12Γ (119887 minus 12)
Γ (119887)
Var (120579) = 119886
2 (119887 minus 1)minus (
119886
2) [Γ (119887 minus 12)
Γ (119887)]
2
119887 gt 1
(16)
With record values 119899 products (or items) take place ontest Consider (119883
119880(1) 119883119880(2) 119883
119880(119899)) is the upper record
values We can obtain that the posterior pdf of 120579 | (119883119880(1)
119883119880(2) 119883
119880(119899)) is given as
1199011(120579 | 119909
119880(1) 119909
119880(119899))
=119871 (120579 | 119909
119880(1) 119909
119880(119899)) 1205871 (120579)
intinfin
0119871 (120579 | 119909
119880(1) 119909
119880(119899)) 1205871 (120579) 119889120579
=
120579minus2(119899+119887)minus1
(1199092
119880(119899)+ 119886)(119899+119887)
119890minus(119909119880(119899)+119886)2120579
2
Γ (119899 + 119887) 2119899+119887minus1
(17)
where 120579 gt 0 and the likelihood function 119871(120579 | 119909119880(1)
119909119880(119899)) as (8)Let 119905 = 1199092
119880(119899)+119886 then the posterior pdf can be rewritten
as
1199011(120579 | 119909
119880(1) 119909
119880(119899)) =
120579minus2(119899+119887)minus1
119905(119899+119887)
119890minus1199052120579
2
Γ (119899 + 119887) 2119899+119887minus1 (18)
and we can show that 1199051205792|119909sim
=(119909119880(1)119909119880(119899))sim 1205942
2(119899+119887)
The Bayesian estimator 120579BS of 120579 based on the squarederror (SE) loss function 119871(120579lowast 120579) = (120579lowast minus120579)2 can be derived asfollows
119864 (120579 | 119909119880(1) 119909
119880(119899))
= int
infin
0
1205791199011(120579 | 119909
119880(1) 119909
119880(119899)) 119889120579
=Γ (119899 + 119887 minus 12)
Γ (119899 + 119887)(119905
2)
12
(19)
where 119905 = 1199092119880(119899)
+ 119886 119899 + 119887 minus 12 gt 0Hence the Bayesian estimator of 120579 based on SE loss
function is given by
120579BS =Γ (119899 + 119887 minus 12)
Γ (119899 + 119887)(119879
2)
12
(20)
where 119879 = 1198832
119880(119899)+ 119886 119899 + 119887 minus 12 gt 0 119886 and 119887 are the
parameters of prior distribution with density as (15)The lifetime performance index 119862
119871of Rayleigh products
can be written as
119862119871= radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579 minusinfin lt 119862
119871lt radic
120587
4 minus 120587 (21)
By using (5) and the Bayesian estimator 120579BS as (20) 119862119871BS
based on the Bayesian estimator 120579BS of 120579 is given by
119862119871BS = radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579BS
= radic120587
4 minus 120587minus radic
2
4 minus 120587sdot 119871 sdot [
Γ (119899 + 119887 minus 12)
Γ (119899 + 119887)(119879
2)
12
]
minus1
(22)
where 119879 = 1198832
119880(119899)+ 119886 119899 + 119887 minus 12 gt 0 119886 and 119887 are the
parameters of prior distribution with density as in (15)We construct a statistical testing procedure to assess
whether the lifetime performance index adheres to therequired level The one-sided credible confidence interval for119862119871are obtained using the pivotal quantity 1199051205792|
119909sim
=(119909119880(1)119909119880(119899))
By using the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899)) given the
specified significance level 120572 the level 100(1minus120572) one-sidedcredible interval for 119862
119871can be derived as follows
Since the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899))sim 1205942
2(119899+119887)
and 12059422(119899+119887)1minus120572
function which represents the lower 1 minus 120572percentile of 1205942
2(119899+119887) then
119875(119905
1205792le 1205942
2(119899+119887)1minus120572| 119909sim) = 1 minus 120572
where 119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899))
997904rArr 119875(119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BS)
timesΓ (119899 + 119887 minus 12)
Γ (119899 + 119887)(1205942
2(119899+119887)1minus120572
2)
12
| 119909sim)
= 1 minus 120572
(23)
where 119862119871as the definition of (5)
From (23) we obtain that a 100(1minus120572)one-sided credibleinterval for 119862
119871is given by
119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BS)
timesΓ (119899 + 119887 minus 12)
Γ (119899 + 119887)(1205942
2(119899+119887)1minus120572
2)
12
(24)
where 119862119871BS is given by (22)
Therefore the 100(1 minus 120572) lower credible interval boundfor 119862119871can be written as
LBBS = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BS)
timesΓ (119899 + 119887 minus 12)
Γ (119899 + 119887)(1205942
2(119899+119887)1minus120572
2)
12
(25)
6 Journal of Applied Mathematics
where 119862119871BS is given by (22) and 119887 is a parameter of prior
distribution with density as (15)The managers can use the one-sided credible interval to
determine whether the product performance adheres to therequired level The proposed testing procedure of 119862
119871with
119862119871BS can be organized as follows
Step 1 Determine the lower lifetime limit 119871 for productsand the performance index value 119888
0 then the testing null
hypothesis 1198670 119862119871le 1198880and the alternative hypothesis
1198671 119862119871gt 1198880are constructed
Step 2 Specify a significance level 120572
Step 3 Given the parameters 119886 and 119887 of prior distributionthe upper record values 119909
sim= (119909
119880(1) 119909119880(2) 119909
119880(119899)) the
lower lifetime limit 119871 and the significance level 120572 then wecan calculate the 100(1 minus 120572) one-sided credible interval[LBBSinfin) for 119862119871 where LBBS as the definition of (25)
Step 4 The decision rule of statistical test is provided asfollows
(1) if the performance index value 1198880notin [LBBSinfin) then
we will reject 1198670 It is concluded that the lifetime
performance index of products meets the requiredlevel
(2) if the performance index value 1198880isin [LBBSinfin) then
we do not reject 1198670 It is concluded that the lifetime
performance index of products does not meet therequired level
52 Testing Procedure for the Lifetime Performance Index 119862119871
with the Bayesian Estimator under Linear Exponential LossFunction Let 119883 be the lifetime of such a product and 119883has a Rayleigh distribution with the pdf as given by (2) Weconsider the linear exponential (LINEX) loss function as (alsosee [18ndash23])
119871 (Δ) prop 119890119888Δminus 119888Δ minus 1 (26)
where Δ = (120579lowast minus 120579) 119888 = 0In this paper we suppose that
Δ1= (
120579lowast
120579)
2
minus 1 (27)
where 120579lowast is an estimator of 120579By using (26)-(27) the posterior expectation of LINEX
loss function 119871(Δ1) is
119864 [119871 (Δ1) | 119909sim] = 119890minus119888119864 [119890119888(120579lowast
120579)2 | 119909sim]
minus 119888119864 [(120579lowast
120579)
2
minus 1 | 119909sim] minus 1
(28)
where 119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899))
The value of 120579lowast that minimizes 119864[119871(Δ1) | 119909sim] denoted by
120579BL is obtained by solving the equation
119889119864 [119871 (Δ1) | 119909sim]
119889120579lowast= 119890minus119888119864[119890119888(120579lowast
120579)2 sdot2119888120579lowast
1205792| 119909sim]
minus 119888119864(2120579lowast
1205792| 119909sim) = 0
(29)
that is 120579BL is the solution to the following equation
119864[120579BL1205792119890119888(120579BL120579)2 | 119909
sim] = 119890119888119864(
120579BL1205792
| 119909sim) (30)
By using (18) and (30) we have
2120579BL (119899 + 119887) sdot119905119899+119887
(119905 minus 21198881205792
BL)119899+119887+1
= 119890119888120579BL
2 (119899 + 119887)
119905
where 119905 = 1199092119880(119899)
+ 119886
(31)
Hence the Bayesian estimator of 120579 under the LINEX lossfunction is given by
120579BL = [119879
2119888(1 minus 119890
minus119888(119899+119887+1))]
12
(32)
where 119879 = 1198832
119880(119899)+ 119886 119886 and 119887 are the parameters of prior
distribution with density as (15)By using (5) and the Bayesian estimator 120579BL as (32) 119862119871BL
based on the Bayesian estimator 120579BL of 120579 is given by
119862119871BL = radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579BL
= radic120587
4 minus 120587minus radic
2
4 minus 120587119871[119879
2119888(1 minus 119890
minus119888(119899+119887+1))]
minus12
(33)
where 119879 = 1198832
119880(119899)+ 119886 119886 and 119887 are the parameters of prior
distribution with density as (15)We construct a statistical testing procedure to assess
whether the lifetime performance index adheres to therequired level The one-sided credible confidence interval for119862119871is obtained using the pivotal quantity 1199051205792|
119909sim
=(119909119880(1)119909119880(119899))
By using the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899)) given the
specified significance level 120572 the level 100(1minus120572) one-sidedcredible interval for 119862
119871can be derived as follows
Journal of Applied Mathematics 7
Since the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899))sim 1205942
2(119899+119887)
and 12059422(119899+119887)1minus120572
function which represents the lower 1 minus 120572percentile of 1205942
2(119899+119887) then
119875(119905
1205792le 1205942
2(119899+119887)1minus120572| 119909sim) = 1 minus 120572
where 119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899))
997904rArr 119875(119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
| 119909sim)
= 1 minus 120572
(34)
where 119862119871as the definition of (5)
From (34) we obtain that a 100(1minus120572)one-sided credibleinterval for 119862
119871is
119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
(35)
where 119862119871BL is given by (33)
Therefore the 100(1 minus 120572) lower credible interval boundfor 119862119871can be written as
LBBL = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
(36)
where 119862119871BL is given by (33) and 119887 is a parameter of prior
distribution with density as (15)The managers can use the one-sided credible interval to
determine whether the product performance attains to therequired level The proposed testing procedure of 119862
119871with
119862119871BL can be organized as follows
Step 1 Determine the lower lifetime limit 119871 for productsand the performance index value 119888
0 then the testing null
hypothesis 1198670 119862119871le 1198880and the alternative hypothesis
1198671 119862119871gt 1198880are constructed
Step 2 Specify a significance level 120572
Step 3 Given the parameters 119886 and 119887 of prior distribution theparameter 119888 of LINEX loss function the upper record values119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899)) the lower lifetime limit 119871 and the
significance level 120572 then we can calculate the 100(1 minus 120572)one-sided credible interval [LBBLinfin) for 119862119871 where LBBL asthe definition of (36)
Step 4 The decision rule of statistical test is provided asfollows
(1) if the performance index value 1198880notin [LBBLinfin) then
we will reject 1198670 It is concluded that the lifetime
performance index of products meets the requiredlevel
(2) if the performance index value 1198880isin [LBBLinfin) then
we do not reject 1198670 It is concluded that the lifetime
performance index of products does not meet therequired level
53 Testing Procedure for the Lifetime Performance Index119862119871with the Bayesian Estimator under General Entropy Loss
Function Let 119883 be the lifetime of such a product and 119883has a Rayleigh distribution with the pdf as given by (2)We consider the general entropy (GE) loss function (also see[12 20])
119871 (120579lowast 120579) prop (
120579lowast
120579)
119902
minus 119902 log(120579lowast
120579) minus 1 (37)
whose minimum occurs at 120579lowast = 120579The loss function is a generalization of entropy loss used
by several authors (eg Dyer and Liu [24] Soliman [25] andSoliman and Elkahlout [26]) where the shape parameter 119902 =1 When 119902 gt 0 a positive error (120579lowast gt 120579) causes more seriousconsequences than a negative error The Bayesian estimator120579BG of 120579 under GE loss function is given by
120579BG = [119864 (120579minus119902| 119909119880(1) 119909
119880(119899))]minus1119902
(38)
By using (18) and (38) then the Bayesian estimator 120579BG of120579 is derived as follows
119864 (120579minus119902| 119909119880(1) 119909
119880(119899))
= int
infin
0
120579minus1199021199011(120579 | 119909
119880(1) 119909
119880(119899)) 119889120579
= (119905
2)
minus1199022
sdotΓ (119899 + 119887 + 1199022)
Γ (119899 + 119887)
(39)
where 119905 = 1199092119880(119899)
+ 119886 Hence the Bayesian estimator 120579BG of 120579under the GE loss function is given by
120579BG = (119879
2)
12
[Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(40)
where 119879 = 1198832
119880(119899)+ 119886 119886 and 119887 are the parameters of prior
distribution with density as (15)By using (5) and the Bayesian estimator 120579BG as (40)119862
119871BGbased on the Bayesian estimator 120579BG of 120579 is given by
119862119871BG = radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579BG
= radic120587
4 minus 120587minus radic
2
4 minus 120587sdot 119871
sdot [(119879
2)
minus1199022Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
1119902
(41)
8 Journal of Applied Mathematics
where 119879 = 1198832
119880(119899)+ 119886 119886 and 119887 are the parameters of prior
distribution with density as (15)We construct a statistical testing procedure to assess
whether the lifetime performance index adheres to therequired level The one-sided credible confidence interval for119862119871is obtained using the pivotal quantity 1199051205792|
119909sim
=(119909119880(1)119909119880(119899))
By using the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899)) given the
specified significance level 120572 the level 100(1minus120572) one-sidedcredible interval for 119862
119871can be derived as follows
Since the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899))sim 1205942
2(119899+119887)
and 12059422(119899+119887)1minus120572
function which represents the lower 1 minus 120572percentile of 1205942
2(119899+119887) then
119875(119905
1205792le 1205942
2(119899+119887)1minus120572| 119909sim) = 1 minus 120572
where 119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899))
997904rArr 119875(119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times[Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
| 119909sim)
= 1 minus 120572
(42)
where 119862119871as the definition of (5)
From (42) we obtain that a 100(1 minus 120572) one-sidedcredible interval for 119862
119871is
119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times [Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
(43)
where 119862119871BG is given by (41)
Therefore the 100(1 minus 120572) lower credible interval boundfor 119862119871can be written as
LBBG = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times [Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
(44)
where 119862119871BG is given by (41) 120572 is the specified significance
level and 119887 is a parameter of prior distribution with densityas (15)
The managers can use the one-sided credible interval todetermine whether the product performance attains to therequired level The proposed testing procedure of 119862
119871with
119862119871BG can be organized as follows
Step 1 Determine the lower lifetime limit 119871 for productsand the performance index value 119888
0 then the testing null
hypothesis 1198670 119862119871le 1198880and the alternative hypothesis
1198671 119862119871gt 1198880are constructed
Step 2 Specify a significance level 120572
Step 3 Given the parameters 119886 and 119887 of prior distributionthe parameter 119902 of GE loss function the upper record values119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899)) the lower lifetime limit 119871 and the
significance level 120572 then we can calculate the 100(1 minus 120572)one-sided credible interval [LBBGinfin) for 119862119871 where LBBG asthe definition of (44)
Step 4 The decision rule of statistical test is provided asfollows
(1) if the performance index value 1198880notin [LBBGinfin) then
we will reject 1198670 It is concluded that the lifetime
performance index of products meets the requiredlevel
(2) if the performance index value 1198880isin [LBBGinfin) then
we do not reject 1198670 It is concluded that the lifetime
performance index of products does not meet therequired level
6 The Monte Carlo Simulation Algorithm ofConfidence (or Credible) Level
In this section we will report the results of a simulationstudy for confidence (or credible) level (1 minus 120572) based on a100(1 minus 120572) one-sided confidence (or credible) interval ofthe lifetime performance index 119862
119871 We considered 120572 = 005
and then generated samples from the Rayleigh distributionwith pdf as in (2) with respect to the record values
(i) The Monte Carlo simulation algorithm of confidencelevel (1 minus 120572) for 119862
119871under MLE is given in the
following steps
Step 1 Given 119899 119886 119887 119871 and 120572 where 119886 gt 0 119887 gt 0
Step 2 For the values of prior parameters (119886 119887) use (15) togenerate 120579 from the square-root inverted-gamma distribu-tion
Step 3(a) The generation of data (119885
1 1198852 119885
119899) is by standard
exponential distribution with parameter 1(b) Set1198842
1= 212057921198851and1198842119894= 21205792119885119894+1198842
119894minus1 for 119894 = 2 119899
1198841= radic21205792119885
1 119884
119894= radic21205792119885
119894+ 1198842
119894minus1 for 119894 = 2 119899
(45)
(1198841 1198842 119884
119899) are the upper record values from the
Rayleigh distribution(c) The value of LBMLE is calculated by
LBMLE = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871MLE)(
1205942
21198991minus120572
2119899)
12
(46)
Journal of Applied Mathematics 9
where 119862119871MLE is given by (11) and 1205942
21198991minus120572function
which represents the lower 1 minus 120572 percentile of 12059422119899
(d) If 119862119871ge LBMLE then count = 1 else count = 0
Step 4
(a) Step 3 is repeated 1000 times(b) The estimation of confidence level (1 minus 120572) is (1 minus 120572) =
(total count)1000
Step 5
(a) Repeat Step 2ndashStep 4 for 100 times then we can getthe 100 estimations of confidence level as follows
(1 minus 120572)1 (1 minus 120572)
2 (1 minus 120572)
100 (47)
(b) The average empirical confidence level 1 minus 120572
of (1 minus 120572)119894 119894 = 1 100 that is 1 minus 120572 =
(1100)sum100
119894=1(1 minus 120572)
119894
(c) The sample mean square error (SMSE) of(1 minus 120572)
1 (1 minus 120572)
2 (1 minus 120572)
100SMSE = (1100)
times sum100
119894=1[(1 minus 120572)
119894minus (1 minus 120572)]
2
The results of simulation are summarized in Table 2 basedon 119871 = 10 the different value of sample size 119899 priorparameter (119886 119887) = (2 5) (6 15) and (2 2) at 120572 = 005respectively The scope of SMSE is between 000413 and000593
Step 1 Given 119899 119886 119887 119871 and 120572 where 119886 gt 0 119887 gt 0
Step 2 For the values of prior parameters (119886 119887) use (15) togenerate 120579 from the square-root inverted-gamma distribu-tion
Step 3
(a) The generation of data (1198851 1198852 119885
119899) is by standard
exponential distribution with parameter 1(b) Set1198842
1= 212057921198851and1198842119894= 21205792119885119894+1198842
119894minus1 for 119894 = 2 119899
1198841= radic21205792119885
1 119884
119894= radic21205792119885
119894+ 1198842
119894minus1 for 119894 = 2 119899
(48)
(1198841 1198842 119884
119899) are the upper record values from the
Rayleigh distribution(c) The values of LBBS LBBL and LBBG are calculated by
(25) (36) and (44) respectively(d) If119862
119871ge LB then count = 1 else count = 0 where LB =
LBBS or LBBL or LBBG
Step 4
(a) Step 3 is repeated 1000 times(b) The estimation of credible level (1 minus 120572) is (1 minus 120572) =
(total count)1000
Step 5
(a) Repeat Step 2ndashStep 4 for 100 times then we can getthe 100 estimations of credible level as follows
(1 minus 120572)1 (1 minus 120572)
2 (1 minus 120572)
100 (49)
(b) The average empirical credible level 1 minus 120572 of (1 minus 120572)119894
119894 = 1 100 that is 1 minus 120572 = (1100)sum100119894=1(1 minus 120572)
119894
(c) The sample mean square error (SMSE) of(1 minus 120572)
1 (1 minus 120572)
2 (1 minus 120572)
100SMSE = (1100)
times sum100
119894=1[(1 minus 120572)
119894minus (1 minus 120572)]
2
Based on 119871 = 10 the different value of sample size119899 prior parameter (119886 119887) = (2 5) (6 15) (2 2) and 120572 =
005 the results of simulation are summarized in Tables 3ndash5under SE loss function LINEX loss function with parameter119888 = minus05 05 15 and GE loss function with parameter119902 = 3 5 8 respectively The following points can be drawn
(a) All of the SMSEs are small enough and the scope ofSMSE is between 000346 and 000467
(b) Fix the size 119899 comparison of prior parameter (119886 119887) =(2 5) (6 15) and (2 2) as follows
(i) the values of SMSEwith prior parameter (119886 119887) =(6 15) and (2 5) are smaller than (119886 119887) = (2 2)respectively
(ii) a comparison of prior parameter (119886 119887) = (2 5)and (2 2) that is fix prior parameter 119886 if priorparameter 119887 increases then it can be seen thatthe SMSE will decrease
(c) In Table 4 fix the size 119899 and the prior parameter (119886 119887)comparison the parameter of LINEX loss function 119888 =minus05 05 15 as followsthe values of SMSE with prior parameter 119888 = 05 15andminus15 are the same as the values of SMSE inTable 3
(d) In Table 5 fix the prior parameter (119886 119887) comparisonof GE loss function parameter 119902 = 3 5 8 as followsthe values of SMSE with prior parameter 119902 = 3 5 and8 are the same as the values of SMSE in Table 3
Hence these results from simulation studies illustrate thatthe performance of our proposed method is acceptableMoreover we suggest that the prior parameter (119886 119887) = (6 15)and (2 5) is appropriate for the square-root inverted-gammadistribution
7 Numerical Example
In this section we propose the new hypothesis testingprocedures to a real-life data In the following examplewe discuss a real-life data for 25 ball bearings to illustratethe use of the new hypothesis testing procedures in thelifetime performance of ball bearings The proposed testingprocedures not only can handle nonnormal lifetime data butalso can handle the upper record values
10 Journal of Applied Mathematics
Table 2 Average empirical confidence level (1 minus 120572) for 119862119871under MLE when 120572 = 005
119899 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
5 093653 (000462) 093482 (000467) 093577 (000507)10 093643 (000449) 093581 (000413) 093450 (000593)15 093523 (000489) 093557 (000503) 093463 (000516)119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 3 Average empirical credible level (1 minus 120572) for 119862119871under SE loss function when 120572 = 005
119899 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
5 094357 (000386) 093760 (000386) 093925 (000395)10 094244 (000359) 093704 (000359) 093664 (000462)15 094096 (000346) 093619 (000426) 093635 (000467)119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 4 Average empirical credible level (1 minus 120572) for 119862119871under LINEX loss function when 120572 = 005
119899 119888 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
minus05 094357 (000386) 093760 (000386) 093925 (000395)5 05 094357 (000386) 093760 (000386) 093925 (000395)
15 094357 (000386) 093760 (000386) 093925 (000395)minus05 094244 (000359) 093704 (000359) 093664 (000462)
10 05 094244 (000359) 093704 (000359) 093664 (000462)15 094244 (000359) 093704 (000359) 093664 (000462)minus05 094096 (000346) 093619 (000426) 093635 (000467)
15 05 094096 (000346) 093619 (000426) 093635 (000467)15 094096 (000346) 093619 (000426) 093635 (000467)
119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 5 Average empirical credible level (1 minus 120572) for 119862119871under GE loss function when 120572 = 005
119899 119902 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
3 094357 (000386) 093760 (000386) 093925 (000395)5 5 094357 (000386) 093760 (000386) 093925 (000395)
8 094357 (000386) 093760 (000386) 093925 (000395)3 094244 (000359) 093704 (000359) 093664 (000462)
10 5 094244 (000359) 093704 (000359) 093664 (000462)8 094244 (000359) 093704 (000359) 093664 (000462)3 094096 (000346) 093619 (000426) 093635 (000467)
15 5 094096 (000346) 093619 (000426) 093635 (000467)8 094096 (000346) 093619 (000426) 093635 (000467)
119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Example The data is the failure times of 25 ball bearings inendurance testThe 25 observations are the number ofmillionrevolutions before failure for each of the ball bearings Thedata come from Caroni [27] as follows
6780 6780 6780 6864 3300 6864 9864 12804 42122892 4560 5184 5556 17340 4848 1788 9312 54124152 5196 12792 8412 10512 10584 and 6888
Raqab and Madi [28] and Lee [29] indicated that a one-parameter Rayleigh distribution is acceptable for these dataIn addition we also have a way to test the hypothesis thatthe failure data come from the Rayleigh distribution Thetesting hypothesis119867
0 119883 comes from a Rayleigh distribution
versus1198671 119883 does not come from a Rayleigh distribution is
constructed under significance level 120572 = 005From a probability plot of Figure 1 by using the Minitab
Statistical Software we can conclude the operational lifetimesdata of 25 ball bearings from the Rayleigh distribution whichis the Weibull distribution with the shape 2 (also see Lee etal [14]) For informative prior we use the prior information119864(120579) = 120579MLE ≐ 54834 and Var(120579) = 02 giving the priorparameter values as (119886 = 6014 119887 = 1001)
Here if only the upper record values have been observedthese are 119909
119880(119894) 119894 = 1 5(= 119899) = 6780 6864 9864
12804 17340
Journal of Applied Mathematics 11
10010
99
90807060504030
20
10
5
32
1
Xi
Probability plot of XiWeibull-95 CI
()
Shape 2
Scale 8002N 25
AD 0431P -value 057
Figure 1 Probability plot for failure times of 25 ball bearings data
(1) The proposed testing procedure of 119862119871with 119862
119871MLE isstated as follows
Step 1 The record values from the above data are (119909119880(119894) 119894 =
1 2 3 4 5) = (6780 6864 9864 12804 17340)
Step 2 The lower specification limit 119871 is assumed to be 2337The deal with the product managersrsquo concerns regardinglifetime performance and the conforming rate 119875
119903of products
is required to exceed 80 percent Referring to Table 1 119862119871is
required to exceed 090Thus the performance index value isset at 119888
0= 090 The testing hypothesis119867
0 119862119871le 090 versus
1198671 119862119871gt 090 is constructed
Step 3 Specify a significance level 120572 = 005
Step 4 We can calculate that the 95 lower confidenceinterval bound for 119862
119871 where
LBMLE = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871MLE)(
1205942
21198991minus120572
2119899)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 126251254)(
1205942
10095
10)
12
= 103285
(50)So the 95 one-sided confidence interval for 119862
119871is
[LBMLEinfin) = [103285infin)
Step 5 Because of the performance index value 1198880= 090 notin
[LBMLEinfin) we reject the null hypothesis1198670 119862119871 le 090
Thus we can conclude that the lifetime performanceindex of data meets the required level
(2) The proposed testing procedures of 119862119871with 119862
119871BS119862119871BL and 119862119871BG are stated as follows
Step 1 The record values from the above data are (119909119880(119894) 119894 =
1 2 3 4 5) = (6780 6864 9864 12804 17340)
Step 2 The lower specification limit 119871 is assumed to be 2337The deal with the product managersrsquo concerns regardinglifetime performance and the conforming rate 119875
119903of products
is required to exceed 80 percent Referring to Table 1 119862119871is
required to exceed 090Thus the performance index value isset at 119888
0= 090 The testing hypothesis119867
0 119862119871le 090 versus
1198671 119862119871gt 090 is constructed
Step 3 Specify a significance level 120572 = 005 the parameter ofLINEX loss function 119888 = 05 and the parameter of GE lossfunction 119902 = 2
Step 4 We can calculate the 95 lower credible intervalbound for 119862
119871 where
LBBS = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BS)
timesΓ (119899 + 119887 minus 12)
Γ (119899 + 119887)(1205942
2(119899+119887)1minus120572
2)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1246074685)
timesΓ (5 + 1001 minus 12)
Γ (5 + 1001)(1205942
2(5+1001)095
2)
12
= 096984
(51)
12 Journal of Applied Mathematics
So the 95 one-sided credible interval for 119862119871
is[LBBSinfin) = [096984infin)
LBBL = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1129561059)
times [1
2 (05)(1 minus 119890
minus05(5+1001+1)) 1205942
2(5+1001)095]
12
= 096984
(52)
So the 95 one-sided credible interval for 119862119871
is[LBBLinfin) = [096984infin)
LBBG = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times [Γ (119899 + 119887 + 1199022)
Τ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1200432998)
times [Γ (5 + 1001 + 22)
Γ (5 + 1001)]
minus12
(1205942
2(5+1001)095
2)
12
= 096984
(53)
So the 95 one-sided credible interval for 119862119871
is[LBBGinfin) = [096984infin)
Step 5 Because of the performance index value 1198880= 090 notin
[LBBSinfin) = [LBBLinfin) = [LBBGinfin) we reject the nullhypothesis119867
0 119862119871le 090
Thus we can conclude that the lifetime performance index ofdata meets the required level In addition by using the priorparameter values (119886 119887) = (6 15) based on the simulationstudies we also obtain that 119888
0= 090 notin [LBBSinfin) =
[LBBLinfin) = [LBBGinfin) = 094033 So we also rejectthe null hypothesis 119867
0 119862119871le 090 Hence the lifetime
performance index of data meets the required level
8 Conclusions
Montgomery [1] proposed a process capability index 119862119871for
larger-the-better quality characteristic The assumption ofmost process capability is normal distribution but it is ofteninvalid In this paper we consider that Rayleigh distributionis the special case of Weibull distribution
Moreover in lifetime data testing experiments the exper-imenter may not always be in a position to observe the life
times of all the products put to test In industry and reliabilitystudies many products fail under stress for example anelectronic component ceases to function in an environmentof too high temperature and a battery dies under the stressof time So in such experiments measurement may be madesequentially and only the record values are observed
In order to let the process capability indices can beeffectively used This study constructs MLE and Bayesianestimator of 119862
119871under assuming the conjugate prior distribu-
tion and SE loss function LINEX loss function and GE lossfunction based on the upper record values from the Rayleighdistribution The Bayesian estimator of 119862
119871is then utilized to
develop a credible interval in the condition of known 119871This study also provides a table of the lifetime perfor-
mance index with its corresponding conforming rate basedon the Rayleigh distributionThat is the conforming rate andthe corresponding 119862
119871can be obtained
Further the Bayesian estimator and posterior distributionof 119862119871are also utilized to construct the testing procedure of
119862119871which is based on a credible interval If you want to test
whether the products meet the required level you can utilizethe proposed testing procedure which is easily applied andan effectively method to test in the condition of known 119871Numerical example is illustrated to show that the proposedtesting procedure is effectively evaluating whether the trueperformance index meets requirements
In addition these results from simulation studies illus-trate that the performance of our proposedmethod is accept-able According to SMSE the Bayesian approach is smallerthan the non-Bayesian approach we suggest that the Bayesianapproach is better than the non-Bayesian approachThen theSMSEs of prior parameters (119886 119887) = (6 15) and (2 5) aresmaller than the SMSE of prior parameter (119886 119887) = (2 2)We suggest that the prior parameters (119886 119887) = (6 15) and(2 5) are appropriate for the square-root inverted-gammadistribution based on the Bayesian estimators under theRayleigh distribution
Acknowledgments
The authors thank the referees for their suggestions andhelpful comments on revising the paper This research waspartially supported by the National Science Council China(Plans nos NSC 101-2221-E-158-003 NSC 99-2118-M-415-001 and NSC 100-2118-M-415-002) Moreover J W Wu andC W Hong have a direct financial relation with the NationalScience Council Taiwan but other authors do not a directfinancial relation with the National Science Council Taiwan
References
[1] D C Montgomery Introduction To Statistical Quality ControlJohn Wiley amp Sons New York NY USA 1985
[2] L I Tong K S Chen and H T Chen ldquoStatistical testingfor assessing the performance of lifetime index of electroniccomponents with exponential distributionrdquo Journal of QualityReliability Management vol 19 pp 812ndash824 2002
[3] C-W Hong J-W Wu and C-H Cheng ldquoComputationalprocedure of performance assessment of lifetime index of
Journal of Applied Mathematics 13
businesses for the Pareto lifetime model with the right type IIcensored samplerdquo Applied Mathematics and Computation vol184 no 2 pp 336ndash350 2007
[4] C W Hong J W Wu and C H Cheng ldquoComputational pro-cedure of performance assessment of lifetime index of Paretolifetime businesses based on confidence intervalrdquo Applied SoftComputing vol 8 no 1 pp 698ndash705 2008
[5] J-W Wu H-M Lee and C-L Lei ldquoComputational testingalgorithmic procedure of assessment for lifetime performanceindex of products with two-parameter exponential distribu-tionrdquo Applied Mathematics and Computation vol 190 no 1 pp116ndash125 2007
[6] W-C Lee J-W Wu and C-W Hong ldquoAssessing the lifetimeperformance index of products with the exponential distribu-tion under progressively type II right censored samplesrdquo Journalof Computational and Applied Mathematics vol 231 no 2 pp648ndash656 2009
[7] W C Lee J W Wu and C L Lei ldquoEvaluating the lifetimeperformance index for the exponential lifetime productsrdquoApplied Mathematical Modelling vol 34 no 5 pp 1217ndash12242010
[8] K N Chandler ldquoThe distribution and frequency of recordvaluesrdquo Journal of the Royal Statistical Society B vol 14 pp 220ndash228 1952
[9] S K Bhattacharya and R K Tyagi ldquoBayesian survival analysisbased on the rayleigh modelrdquo Trabajos de Estadistica vol 5 no1 pp 81ndash92 1990
[10] A D Cliff and J K Ord ldquoModel building and the analysis ofspatial pattern in human geography (with discussion)rdquo Journalof the Royal Statistical Society B vol 37 no 3 pp 297ndash348 1975
[11] D D Dyer and C W Whisenand ldquoBest linear unbiasedestimator of the parameter of the Rayleigh distribution I Smallsample theory for censored order statisticsrdquo IEEE Transactionson Reliability vol R-22 no 1 pp 27ndash34 1973
[12] A A Soliman and F M Al-Aboud ldquoBayesian inference usingrecord values from Rayleigh model with applicationrdquo EuropeanJournal of Operational Research vol 185 no 2 pp 659ndash6722008
[13] StatSoft STATISTICA Software Version 7 StatSoft Tulsa Okla2005
[14] W-C Lee J-W Wu M-L Hong L-S Lin and R-L ChanldquoAssessing the lifetime performance index of Rayleigh productsbased on the Bayesian estimation under progressive type IIright censored samplesrdquo Journal of Computational and AppliedMathematics vol 235 no 6 pp 1676ndash1688 2011
[15] M Ahsanullah Record Statistics Nova Science New York NYUSA 1995
[16] B C Arnold N Balakrishnan and H N Nagaraja RecordsJohn Wiley amp Sons New York NY USA 1998
[17] P W Zehna ldquoInvariance of maximum likelihood estimatorsrdquoAnnals of Mathematical Statistics vol 37 p 744 1966
[18] A Zellner ldquoBayesian estimation and prediction using asymmet-ric loss functionsrdquo Journal of the American Statistical Associa-tion vol 81 no 394 pp 446ndash451 1986
[19] S-Y Huang ldquoEmpirical Bayes testing procedures in somenonexponential families using asymmetric linex loss functionrdquoJournal of Statistical Planning and Inference vol 46 no 3 pp293ndash309 1995
[20] R Calabria andG Pulcini ldquoPoint estimation under asymmetricloss functions for left-truncated exponential samplesrdquo Commu-nications in StatisticsmdashTheory and Methods vol 25 no 3 pp585ndash600 1996
[21] A T K Wan G Zou and A H Lee ldquoMinimax and Γ-minimaxestimation for the Poisson distribution under LINEX loss whenthe parameter space is restrictedrdquo Statistics and ProbabilityLetters vol 50 no 1 pp 23ndash32 2000
[22] S Anatolyev ldquoKernel estimation under linear-exponential lossrdquoEconomics Letters vol 91 no 1 pp 39ndash43 2006
[23] X Li Y Shi J Wei and J Chai ldquoEmpirical Bayes estimators ofreliability performances using LINEX loss under progressivelytype-II censored samplesrdquo Mathematics and Computers inSimulation vol 73 no 5 pp 320ndash326 2007
[24] D K Dyer and P S L Liu ldquoOn comparison of estimators in ageneralized life modelrdquo Microelectronics Reliability vol 32 no1-2 pp 207ndash221 1992
[25] A A Soliman ldquoEstimation of parameters of life from progres-sively censored data using Burr-XII modelrdquo IEEE Transactionson Reliability vol 54 no 1 pp 34ndash42 2005
[26] A A Soliman and G R Elkahlout ldquoBayes estimation of thelogistic distribution based on progressively censored samplesrdquoJournal of Applied Statistical Science vol 14 no 3-4 pp 281ndash2932005
[27] C Caroni ldquoThe correct ldquoball bearingsrdquo datardquo Lifetime DataAnalysis vol 8 no 4 pp 395ndash399 2002
[28] M Z Raqab and M T Madi ldquoBayesian prediction of the totaltime on test using doubly censored Rayleigh datardquo Journal ofStatistical Computation and Simulation vol 72 no 10 pp 781ndash789 2002
[29] W C Lee ldquoStatistical testing for assessing lifetime performanceIndex of the Rayleigh lifetime productsrdquo Journal of the ChineseInstitute of Industrial Engineers vol 25 pp 433ndash445 2008
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
2 Journal of Applied Mathematics
breaking stress or failure point varies even among identicalitems Hence in such experiments measurements may bemade sequentially and only the record values (lower or upper)are observed Record values arise naturally in many real-life applications involving data relating to weather sportseconomics and life tests According to themodel of Chandler[8] there are some situations in lifetime testing experimentswhere the failure time of a product is recorded if it exceedsall preceding failure times These recorded failure times arethe upper record value sequence In general let 119883
1 1198832
be a sequence of independent and identically distributed(iid) random variables having the same distribution asthe (population) random variable 119883 with the cumulativedistribution function (cdf) 119865(119909) and the probability densityfunction (pdf) 119891(119909) An observation 119883
119895will be called an
upper record value if it exceeds in value all of the precedingobservations that is if119883
119895gt 119883119894 for every 119894 lt 119895 The sequence
of record times 119879119899 119899 ge 1 is defined as follows
Let 1198791= 1 with probability 1 for 119899 ge 2 119879
119899= min119895
119883119895gt 119883119879119899minus1
A sequence of upper record values 119883119880(119899) is
then defined by
119883119880(119899)
= 119883119879119899
119899 = 1 2 (1)
Let 119883sim= 119883119880(1) 119883119880(2) 119883
119880(119899) be the first 119899 upper record
values arising from a sequence of iid random variables1198831 1198832 with cdf 119865(119909)
In this study we consider the case of the upper recordvalues instead of complete data or censored data In orderto evaluate the lifetime performance of nonnormal datawith upper record values the lifetime performance index 119862
119871
is utilized to measure product quality under the Rayleighdistribution with the upper record values The Rayleighdistribution is a nonnormal distribution and a special case ofthe Weibull distribution which provides a population modeluseful in several areas of statistics including life testing andreliability whose age with time as its failure rate is a linearfunction of time Bhattacharya and Tyagi [9] mentioned thatin some clinical studies dealing with cancer patients thesurvival pattern follows the Rayleigh distribution Cliff andOrd [10] showed that the Rayleigh distribution arises as thedistribution of the distance between an individual and itsnearest neighbor when the special pattern is generated by thePoisson process Dyer and Whisenand [11] demonstrated theimportance of this distribution in communication engineer-ing (also see Soliman and Al-Aboud [12])The pdf and cdfof the Rayleigh distribution are given respectively by
119891 (119909) =119909
1205792119890minus1199092
21205792
(2)
119865 (119909) = 1 minus 119890minus1199092
21205792
(3)
where 119909 gt 0 and 120579 gt 0 respectively When119883 comes from theRayleigh distribution the mean of 119883 is 119864(119883) = radic1205872120579 andthe standard deviation of119883 is 120590 = radic(4 minus 120587)2120579
Bayesian and non-Bayesian approaches have been used toobtain the estimators of the parameter 120579 under the Rayleigh
distribution with the upper record values Soliman and Al-Aboud [12] compared the performance of the Bayesian esti-mators with non-Bayesian estimators such as the MLE andthe best linear unbiased (BLUE) estimatorThe Bayesian esti-mators are developed under symmetric and nonsymmetricloss functions The symmetric loss function is squared-error(SE) loss function The nonsymmetric loss function includeslinear exponential (LINEX) and general entropy (GE) lossfunctions In recent decades the Bayesian viewpoint hasreceived frequent attention for analyzing failure data andother time-to-event data and has been often proposed as avalid alternative to traditional statistical perspectives TheBayesian approach to estimation of the parameters andreliability analysis allows prior subjective knowledge onlifetime parameters and technical information on the failuremechanism as well as experimental data Bayesian methodsusually require less sample data to achieve the same qualityof inferences than methods based on sampling theory (alsosee [12])
The main aim of this study will construct the Bayesianestimator of 119862
119871under a Rayleigh distribution with upper
record values The Bayesian estimators of 119862119871are developed
under symmetric and nonsymmetric loss functions Theestimators of 119862
119871are then utilized to develop a credible
interval respectively The new testing procedures of credibleinterval for 119862
119871can be employed by managers to assess
whether the products performance adheres to the requiredlevel in the condition of known 119871 The new proposed testingprocedures can handle nonnormal lifetime data with upperrecord values Moreover we will evaluate the performance ofthe new proposed testing procedures with Bayesian and non-Bayesian approaches
The rest of this study is organized as follows Section 2introduces some properties of the lifetime performanceindex for lifetime of product with the Rayleigh distributionSection 3 discusses the relationship between the lifetimeperformance index and conforming rate Section 4 developsa hypothesis testing procedure for the lifetime performanceindex119862
119871with the non-Bayesian approach Section 5 develops
a hypothesis testing procedure for the lifetime performanceindex 119862
119871with the Bayesian approach Section 6 discusses the
MonteCarlo simulation algorithmof confidence (or credible)level One numerical example and concluding remarks aremade in Sections 7 and 8 respectively
2 The Lifetime Performance Index
Montgomery [1] has developed a process capability index 119862119871
to measure the larger-the-better quality characteristic Then119862119871is defined by
119862119871=120583 minus 119871
120590 (4)
where 120583 120590 and 119871 are the process mean the process standarddeviation and the lower specification limit respectively
To assess the product performance of products 119862119871can
be defined as the lifetime performance index If 119883 comesfrom theRayleigh distribution then the lifetime performance
Journal of Applied Mathematics 3
index 119862119871can be rewritten as
119862119871=120583 minus 119871
120590=radic1205872120579 minus 119871
radic(4 minus 120587) 2120579= radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579
minusinfin lt 119862119871lt radic
120587
4 minus 120587
(5)
where 120583 = 119864(119883) = radic1205872120579 is the process mean 120590 =
radicVar(119883) = radic(4 minus 120587)2120579 is the process standard deviationand 119871 is the lower specification limit
3 The Conforming Rate
If the lifetime of a product 119883 exceeds the lower specificationlimit 119871 then the product is defined as a conforming productThe ratio of conforming products is known as the conformingrate and can be defined as
119875119903= 119875 (119883 ge 119871) = 119890
minus(12)(radic1205872minusradic(4minus120587)2119862119871)2
minus infin lt 119862119871lt radic120587 (4 minus 120587)
(6)
Obviously a strictly increasing relationship existsbetween the conforming rate 119875
119903and the lifetime
performance index 119862119871 Table 1 lists various 119862
119871values and
the corresponding conforming rate 119875119903by using STATISTICA
software [13] (also see [14])For the 119862
119871values which are not listed in Table 1 the
conforming rate 119875119903can be obtained through interpolation In
addition since a one-to-one mathematical relationship existsbetween the conforming rate119875
119903and the lifetime performance
index 119862119871 Therefore utilizing the one-to-one relationship
between 119875119903and 119862
119871 lifetime performance index can be
a flexible and effective tool not only evaluating productperformance but also estimating the conforming rate 119875
119903
4 Testing Procedure for the LifetimePerformance Index 119862
119871with
Non-Bayesian Approach
This section will apply non-Bayesian approach to construct amaximum likelihood estimator (MLE) of119862
119871under aRayleigh
distribution with upper record values The MLE of 119862119871is
then utilized to develop a new hypothesis testing procedurein the condition of known 119871 Assuming that the requiredindex value of lifetime performance is larger than 119888
0 where
1198880denotes the target value the null hypothesis 119867
0 119862119871le 1198880
and the alternative hypothesis1198671 119862119871gt 1198880are constructed
Based on the new hypothesis testing procedure the lifetimeperformance of products is easy to assess
Let 119883 be the lifetime of such a product and 119883 hasa Rayleigh distribution with the pdf as given by (2)Let 119883sim
= (119883119880(1) 119883119880(2) 119883
119880(119899)) be the first 119899 upper
record values arising from a sequence of iid Rayleigh
variables with pdf as given by (2) Since the joint pdf of(119883119880(1) 119883119880(2) 119883
119880(119899)) is
119891 (119909119880(119899))
119899minus1
prod
119894=1
119891 (119909119880(119894))
1 minus 119865 (119909119880(119894)) (7)
where 119891(119909) and 119865(119909) are the pdf and cdf of 119883 respec-tively (also see [12 15 16]) So the likelihood function of(119883119880(1) 119883119880(2) 119883
119880(119899)) is given as
119871 (120579 | 119909119880(1) 119909
119880(119899)) =
119899
prod
119894=1
119909119880(119894)
1205792 exp(minus
1199092
119880(119899)
21205792) (8)
The natural logarithm of the likelihood function with (8) is
ln 119871 (120579 | 119909119880(1) 119909
119880(119899)) =
119899
sum
119894=1
ln (119909119880(119894)) minus 2119899 ln 120579 minus
1199092
119880(119899)
21205792
(9)
Upon differentiating (9) with respect to 120579 and equating theresult to zero theMLE of the parameter 120579 can be shown to be
120579MLE =119883119880(119899)
radic2119899(10)
(also see [12]) By using the invariance of MLE (see Zehna[17]) the MLE of 119862
119871can be written as given by
119862119871MLE = radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579MLE
= radic120587
4 minus 120587minus radic
2
4 minus 120587sdot 119871 sdot
radic2119899
119883119880(119899)
(11)
Construct a statistical testing procedure to assess whetherthe lifetime performance index adheres to the required levelThe one-sided confidence interval for 119862
119871is obtained using
the pivotal quantity 1198832119880(119899)1205792 By using the pivotal quantity
1198832
119880(119899)1205792 given the specified significance level 120572 the level
100(1 minus 120572) one-sided confidence interval for 119862119871can be
derived as followsSince the pivotal quantity 1198832
119880(119899)1205792sim 1205942
2119899 and 1205942
21198991minus120572
function which represents the lower 1 minus 120572 percentile of 12059422119899
then
119875(1198832
119880(119899)
1205792le 1205942
21198991minus120572) = 1 minus 120572
997904rArr 119875(119862119871ge radic
120587
4 minus 120587minus(radic
120587
4 minus 120587minus119862119871MLE)(
1205942
21198991minus120572
2119899)
12
)
= 1 minus 120572
(12)
where 119862119871= radic120587(4 minus 120587) minus radic2(4 minus 120587)(119871120579) From (12) we
4 Journal of Applied Mathematics
Table 1 The lifetime performance index 119862119871versus the conforming rate 119875
119903
119862119871
119875119903
119862119871
119875119903
119862119871
119875119903
minusinfin 0000000 040 0611832 120 0896627minus450 0000147 045 0631685 125 0909965minus400 0000551 050 0651484 130 0922511minus350 0001858 055 0671182 135 0934227minus300 0005628 060 0690734 140 0945076minus250 0015308 065 0710094 145 0955027minus200 0037404 070 0729213 150 0964047minus150 0082094 075 0748044 155 0972109minus100 0161849 080 0766539 160 0979187minus050 0286621 085 0784648 165 0985259000 0455938 090 0802324 170 0990306015 0513214 095 0819518 175 0994310020 0532718 100 0836183 180 0997261025 0552370 105 0852271 185 0999147030 0572132 110 0867738 190 0999963035 0591967 115 0882538 191 0999997
obtain that a 100(1 minus 120572) one-sided confidence interval for119862119871is
119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871MLE)(
1205942
21198991minus120572
2119899)
12
(13)
where119862119871MLE is given by (11)Therefore the 100(1minus120572) lower
confidence interval bound for 119862119871can be written as
LBMLE = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871MLE)(
1205942
21198991minus120572
2119899)
12
(14)
where 119862119871MLE 120572 and 119899 denote the MLE of 119862
119871 the specified
significance level and the upper record sample of sizerespectively
The managers can use the one-sided confidence intervalto determine whether the product performance adheres tothe required levelThe proposed testing procedure of119862
119871with
119862119871MLE can be organized as follows
Step 1 Determine the lower lifetime limit 119871 for productsand the performance index value 119888
0 then the testing null
hypothesis 1198670 119862119871le 1198880and the alternative hypothesis
1198671 119862119871gt 1198880are constructed
Step 2 Specify a significance level 120572
Step 3 Given the upper record values 119909sim= (119909119880(1) 119909119880(2)
119909119880(119899)) the lower lifetime limit 119871 and the significance level 120572
then we can calculate the 100(1 minus 120572) one-sided confidenceinterval [LBMLEinfin) for 119862119871 where LBMLE as the definition of(14)
Step 4 The decision rule of statistical test is provided asfollows
(1) if the performance index value 1198880notin [LBMLEinfin) then
we will reject 1198670 It is concluded that the lifetime
performance index of products meets the requiredlevel
(2) if the performance index value 1198880isin [LBMLEinfin) then
we do not reject 1198670 It is concluded that the lifetime
performance index of products does not meet therequired level
5 Testing Procedure forthe Lifetime Performance Index 119862
119871with
the Bayesian Approach
This section will apply the Bayesian approach to constructan estimator of 119862
119871under a Rayleigh distribution with upper
record values The Bayesian estimators are developed undersymmetric and nonsymmetric loss functionsThe symmetricloss function is squared-error (SE) loss function The non-symmetric loss function includes linear exponential (LINEX)and general entropy (GE) loss functions The Bayesian esti-mator of 119862
119871is then utilized to develop a new hypothesis
testing procedure in the condition of known 119871 Assumingthat the required index value of lifetime performance is largerthan 119888
0 where 119888
0denotes the target value the null hypothesis
1198670 119862119871le 1198880and the alternative hypothesis119867
1 119862119871gt 1198880are
constructed Based on the new hypothesis testing procedurethe lifetime performance of products is easy to assess
51 Testing Procedure for the Lifetime Performance Index119862119871with the Bayesian Estimator under Squared-Error Loss
Function Let 119883 be the lifetime of such a product and 119883has a Rayleigh distribution with the pdf as given by (2)We consider the conjugate prior distribution of the formwhich is defined by the square-root inverted-gamma density
Journal of Applied Mathematics 5
as follows
1205871 (120579) =
119886119887
Γ (119887) 2119887minus1120579minus2119887minus1
119890minus1198862120579
2
(15)
where 120579 gt 0 119886 gt 0 and 119887 gt 0
119864 (120579) = (119886
2)
12Γ (119887 minus 12)
Γ (119887)
Var (120579) = 119886
2 (119887 minus 1)minus (
119886
2) [Γ (119887 minus 12)
Γ (119887)]
2
119887 gt 1
(16)
With record values 119899 products (or items) take place ontest Consider (119883
119880(1) 119883119880(2) 119883
119880(119899)) is the upper record
values We can obtain that the posterior pdf of 120579 | (119883119880(1)
119883119880(2) 119883
119880(119899)) is given as
1199011(120579 | 119909
119880(1) 119909
119880(119899))
=119871 (120579 | 119909
119880(1) 119909
119880(119899)) 1205871 (120579)
intinfin
0119871 (120579 | 119909
119880(1) 119909
119880(119899)) 1205871 (120579) 119889120579
=
120579minus2(119899+119887)minus1
(1199092
119880(119899)+ 119886)(119899+119887)
119890minus(119909119880(119899)+119886)2120579
2
Γ (119899 + 119887) 2119899+119887minus1
(17)
where 120579 gt 0 and the likelihood function 119871(120579 | 119909119880(1)
119909119880(119899)) as (8)Let 119905 = 1199092
119880(119899)+119886 then the posterior pdf can be rewritten
as
1199011(120579 | 119909
119880(1) 119909
119880(119899)) =
120579minus2(119899+119887)minus1
119905(119899+119887)
119890minus1199052120579
2
Γ (119899 + 119887) 2119899+119887minus1 (18)
and we can show that 1199051205792|119909sim
=(119909119880(1)119909119880(119899))sim 1205942
2(119899+119887)
The Bayesian estimator 120579BS of 120579 based on the squarederror (SE) loss function 119871(120579lowast 120579) = (120579lowast minus120579)2 can be derived asfollows
119864 (120579 | 119909119880(1) 119909
119880(119899))
= int
infin
0
1205791199011(120579 | 119909
119880(1) 119909
119880(119899)) 119889120579
=Γ (119899 + 119887 minus 12)
Γ (119899 + 119887)(119905
2)
12
(19)
where 119905 = 1199092119880(119899)
+ 119886 119899 + 119887 minus 12 gt 0Hence the Bayesian estimator of 120579 based on SE loss
function is given by
120579BS =Γ (119899 + 119887 minus 12)
Γ (119899 + 119887)(119879
2)
12
(20)
where 119879 = 1198832
119880(119899)+ 119886 119899 + 119887 minus 12 gt 0 119886 and 119887 are the
parameters of prior distribution with density as (15)The lifetime performance index 119862
119871of Rayleigh products
can be written as
119862119871= radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579 minusinfin lt 119862
119871lt radic
120587
4 minus 120587 (21)
By using (5) and the Bayesian estimator 120579BS as (20) 119862119871BS
based on the Bayesian estimator 120579BS of 120579 is given by
119862119871BS = radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579BS
= radic120587
4 minus 120587minus radic
2
4 minus 120587sdot 119871 sdot [
Γ (119899 + 119887 minus 12)
Γ (119899 + 119887)(119879
2)
12
]
minus1
(22)
where 119879 = 1198832
119880(119899)+ 119886 119899 + 119887 minus 12 gt 0 119886 and 119887 are the
parameters of prior distribution with density as in (15)We construct a statistical testing procedure to assess
whether the lifetime performance index adheres to therequired level The one-sided credible confidence interval for119862119871are obtained using the pivotal quantity 1199051205792|
119909sim
=(119909119880(1)119909119880(119899))
By using the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899)) given the
specified significance level 120572 the level 100(1minus120572) one-sidedcredible interval for 119862
119871can be derived as follows
Since the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899))sim 1205942
2(119899+119887)
and 12059422(119899+119887)1minus120572
function which represents the lower 1 minus 120572percentile of 1205942
2(119899+119887) then
119875(119905
1205792le 1205942
2(119899+119887)1minus120572| 119909sim) = 1 minus 120572
where 119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899))
997904rArr 119875(119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BS)
timesΓ (119899 + 119887 minus 12)
Γ (119899 + 119887)(1205942
2(119899+119887)1minus120572
2)
12
| 119909sim)
= 1 minus 120572
(23)
where 119862119871as the definition of (5)
From (23) we obtain that a 100(1minus120572)one-sided credibleinterval for 119862
119871is given by
119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BS)
timesΓ (119899 + 119887 minus 12)
Γ (119899 + 119887)(1205942
2(119899+119887)1minus120572
2)
12
(24)
where 119862119871BS is given by (22)
Therefore the 100(1 minus 120572) lower credible interval boundfor 119862119871can be written as
LBBS = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BS)
timesΓ (119899 + 119887 minus 12)
Γ (119899 + 119887)(1205942
2(119899+119887)1minus120572
2)
12
(25)
6 Journal of Applied Mathematics
where 119862119871BS is given by (22) and 119887 is a parameter of prior
distribution with density as (15)The managers can use the one-sided credible interval to
determine whether the product performance adheres to therequired level The proposed testing procedure of 119862
119871with
119862119871BS can be organized as follows
Step 1 Determine the lower lifetime limit 119871 for productsand the performance index value 119888
0 then the testing null
hypothesis 1198670 119862119871le 1198880and the alternative hypothesis
1198671 119862119871gt 1198880are constructed
Step 2 Specify a significance level 120572
Step 3 Given the parameters 119886 and 119887 of prior distributionthe upper record values 119909
sim= (119909
119880(1) 119909119880(2) 119909
119880(119899)) the
lower lifetime limit 119871 and the significance level 120572 then wecan calculate the 100(1 minus 120572) one-sided credible interval[LBBSinfin) for 119862119871 where LBBS as the definition of (25)
Step 4 The decision rule of statistical test is provided asfollows
(1) if the performance index value 1198880notin [LBBSinfin) then
we will reject 1198670 It is concluded that the lifetime
performance index of products meets the requiredlevel
(2) if the performance index value 1198880isin [LBBSinfin) then
we do not reject 1198670 It is concluded that the lifetime
performance index of products does not meet therequired level
52 Testing Procedure for the Lifetime Performance Index 119862119871
with the Bayesian Estimator under Linear Exponential LossFunction Let 119883 be the lifetime of such a product and 119883has a Rayleigh distribution with the pdf as given by (2) Weconsider the linear exponential (LINEX) loss function as (alsosee [18ndash23])
119871 (Δ) prop 119890119888Δminus 119888Δ minus 1 (26)
where Δ = (120579lowast minus 120579) 119888 = 0In this paper we suppose that
Δ1= (
120579lowast
120579)
2
minus 1 (27)
where 120579lowast is an estimator of 120579By using (26)-(27) the posterior expectation of LINEX
loss function 119871(Δ1) is
119864 [119871 (Δ1) | 119909sim] = 119890minus119888119864 [119890119888(120579lowast
120579)2 | 119909sim]
minus 119888119864 [(120579lowast
120579)
2
minus 1 | 119909sim] minus 1
(28)
where 119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899))
The value of 120579lowast that minimizes 119864[119871(Δ1) | 119909sim] denoted by
120579BL is obtained by solving the equation
119889119864 [119871 (Δ1) | 119909sim]
119889120579lowast= 119890minus119888119864[119890119888(120579lowast
120579)2 sdot2119888120579lowast
1205792| 119909sim]
minus 119888119864(2120579lowast
1205792| 119909sim) = 0
(29)
that is 120579BL is the solution to the following equation
119864[120579BL1205792119890119888(120579BL120579)2 | 119909
sim] = 119890119888119864(
120579BL1205792
| 119909sim) (30)
By using (18) and (30) we have
2120579BL (119899 + 119887) sdot119905119899+119887
(119905 minus 21198881205792
BL)119899+119887+1
= 119890119888120579BL
2 (119899 + 119887)
119905
where 119905 = 1199092119880(119899)
+ 119886
(31)
Hence the Bayesian estimator of 120579 under the LINEX lossfunction is given by
120579BL = [119879
2119888(1 minus 119890
minus119888(119899+119887+1))]
12
(32)
where 119879 = 1198832
119880(119899)+ 119886 119886 and 119887 are the parameters of prior
distribution with density as (15)By using (5) and the Bayesian estimator 120579BL as (32) 119862119871BL
based on the Bayesian estimator 120579BL of 120579 is given by
119862119871BL = radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579BL
= radic120587
4 minus 120587minus radic
2
4 minus 120587119871[119879
2119888(1 minus 119890
minus119888(119899+119887+1))]
minus12
(33)
where 119879 = 1198832
119880(119899)+ 119886 119886 and 119887 are the parameters of prior
distribution with density as (15)We construct a statistical testing procedure to assess
whether the lifetime performance index adheres to therequired level The one-sided credible confidence interval for119862119871is obtained using the pivotal quantity 1199051205792|
119909sim
=(119909119880(1)119909119880(119899))
By using the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899)) given the
specified significance level 120572 the level 100(1minus120572) one-sidedcredible interval for 119862
119871can be derived as follows
Journal of Applied Mathematics 7
Since the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899))sim 1205942
2(119899+119887)
and 12059422(119899+119887)1minus120572
function which represents the lower 1 minus 120572percentile of 1205942
2(119899+119887) then
119875(119905
1205792le 1205942
2(119899+119887)1minus120572| 119909sim) = 1 minus 120572
where 119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899))
997904rArr 119875(119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
| 119909sim)
= 1 minus 120572
(34)
where 119862119871as the definition of (5)
From (34) we obtain that a 100(1minus120572)one-sided credibleinterval for 119862
119871is
119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
(35)
where 119862119871BL is given by (33)
Therefore the 100(1 minus 120572) lower credible interval boundfor 119862119871can be written as
LBBL = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
(36)
where 119862119871BL is given by (33) and 119887 is a parameter of prior
distribution with density as (15)The managers can use the one-sided credible interval to
determine whether the product performance attains to therequired level The proposed testing procedure of 119862
119871with
119862119871BL can be organized as follows
Step 1 Determine the lower lifetime limit 119871 for productsand the performance index value 119888
0 then the testing null
hypothesis 1198670 119862119871le 1198880and the alternative hypothesis
1198671 119862119871gt 1198880are constructed
Step 2 Specify a significance level 120572
Step 3 Given the parameters 119886 and 119887 of prior distribution theparameter 119888 of LINEX loss function the upper record values119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899)) the lower lifetime limit 119871 and the
significance level 120572 then we can calculate the 100(1 minus 120572)one-sided credible interval [LBBLinfin) for 119862119871 where LBBL asthe definition of (36)
Step 4 The decision rule of statistical test is provided asfollows
(1) if the performance index value 1198880notin [LBBLinfin) then
we will reject 1198670 It is concluded that the lifetime
performance index of products meets the requiredlevel
(2) if the performance index value 1198880isin [LBBLinfin) then
we do not reject 1198670 It is concluded that the lifetime
performance index of products does not meet therequired level
53 Testing Procedure for the Lifetime Performance Index119862119871with the Bayesian Estimator under General Entropy Loss
Function Let 119883 be the lifetime of such a product and 119883has a Rayleigh distribution with the pdf as given by (2)We consider the general entropy (GE) loss function (also see[12 20])
119871 (120579lowast 120579) prop (
120579lowast
120579)
119902
minus 119902 log(120579lowast
120579) minus 1 (37)
whose minimum occurs at 120579lowast = 120579The loss function is a generalization of entropy loss used
by several authors (eg Dyer and Liu [24] Soliman [25] andSoliman and Elkahlout [26]) where the shape parameter 119902 =1 When 119902 gt 0 a positive error (120579lowast gt 120579) causes more seriousconsequences than a negative error The Bayesian estimator120579BG of 120579 under GE loss function is given by
120579BG = [119864 (120579minus119902| 119909119880(1) 119909
119880(119899))]minus1119902
(38)
By using (18) and (38) then the Bayesian estimator 120579BG of120579 is derived as follows
119864 (120579minus119902| 119909119880(1) 119909
119880(119899))
= int
infin
0
120579minus1199021199011(120579 | 119909
119880(1) 119909
119880(119899)) 119889120579
= (119905
2)
minus1199022
sdotΓ (119899 + 119887 + 1199022)
Γ (119899 + 119887)
(39)
where 119905 = 1199092119880(119899)
+ 119886 Hence the Bayesian estimator 120579BG of 120579under the GE loss function is given by
120579BG = (119879
2)
12
[Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(40)
where 119879 = 1198832
119880(119899)+ 119886 119886 and 119887 are the parameters of prior
distribution with density as (15)By using (5) and the Bayesian estimator 120579BG as (40)119862
119871BGbased on the Bayesian estimator 120579BG of 120579 is given by
119862119871BG = radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579BG
= radic120587
4 minus 120587minus radic
2
4 minus 120587sdot 119871
sdot [(119879
2)
minus1199022Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
1119902
(41)
8 Journal of Applied Mathematics
where 119879 = 1198832
119880(119899)+ 119886 119886 and 119887 are the parameters of prior
distribution with density as (15)We construct a statistical testing procedure to assess
whether the lifetime performance index adheres to therequired level The one-sided credible confidence interval for119862119871is obtained using the pivotal quantity 1199051205792|
119909sim
=(119909119880(1)119909119880(119899))
By using the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899)) given the
specified significance level 120572 the level 100(1minus120572) one-sidedcredible interval for 119862
119871can be derived as follows
Since the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899))sim 1205942
2(119899+119887)
and 12059422(119899+119887)1minus120572
function which represents the lower 1 minus 120572percentile of 1205942
2(119899+119887) then
119875(119905
1205792le 1205942
2(119899+119887)1minus120572| 119909sim) = 1 minus 120572
where 119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899))
997904rArr 119875(119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times[Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
| 119909sim)
= 1 minus 120572
(42)
where 119862119871as the definition of (5)
From (42) we obtain that a 100(1 minus 120572) one-sidedcredible interval for 119862
119871is
119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times [Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
(43)
where 119862119871BG is given by (41)
Therefore the 100(1 minus 120572) lower credible interval boundfor 119862119871can be written as
LBBG = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times [Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
(44)
where 119862119871BG is given by (41) 120572 is the specified significance
level and 119887 is a parameter of prior distribution with densityas (15)
The managers can use the one-sided credible interval todetermine whether the product performance attains to therequired level The proposed testing procedure of 119862
119871with
119862119871BG can be organized as follows
Step 1 Determine the lower lifetime limit 119871 for productsand the performance index value 119888
0 then the testing null
hypothesis 1198670 119862119871le 1198880and the alternative hypothesis
1198671 119862119871gt 1198880are constructed
Step 2 Specify a significance level 120572
Step 3 Given the parameters 119886 and 119887 of prior distributionthe parameter 119902 of GE loss function the upper record values119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899)) the lower lifetime limit 119871 and the
significance level 120572 then we can calculate the 100(1 minus 120572)one-sided credible interval [LBBGinfin) for 119862119871 where LBBG asthe definition of (44)
Step 4 The decision rule of statistical test is provided asfollows
(1) if the performance index value 1198880notin [LBBGinfin) then
we will reject 1198670 It is concluded that the lifetime
performance index of products meets the requiredlevel
(2) if the performance index value 1198880isin [LBBGinfin) then
we do not reject 1198670 It is concluded that the lifetime
performance index of products does not meet therequired level
6 The Monte Carlo Simulation Algorithm ofConfidence (or Credible) Level
In this section we will report the results of a simulationstudy for confidence (or credible) level (1 minus 120572) based on a100(1 minus 120572) one-sided confidence (or credible) interval ofthe lifetime performance index 119862
119871 We considered 120572 = 005
and then generated samples from the Rayleigh distributionwith pdf as in (2) with respect to the record values
(i) The Monte Carlo simulation algorithm of confidencelevel (1 minus 120572) for 119862
119871under MLE is given in the
following steps
Step 1 Given 119899 119886 119887 119871 and 120572 where 119886 gt 0 119887 gt 0
Step 2 For the values of prior parameters (119886 119887) use (15) togenerate 120579 from the square-root inverted-gamma distribu-tion
Step 3(a) The generation of data (119885
1 1198852 119885
119899) is by standard
exponential distribution with parameter 1(b) Set1198842
1= 212057921198851and1198842119894= 21205792119885119894+1198842
119894minus1 for 119894 = 2 119899
1198841= radic21205792119885
1 119884
119894= radic21205792119885
119894+ 1198842
119894minus1 for 119894 = 2 119899
(45)
(1198841 1198842 119884
119899) are the upper record values from the
Rayleigh distribution(c) The value of LBMLE is calculated by
LBMLE = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871MLE)(
1205942
21198991minus120572
2119899)
12
(46)
Journal of Applied Mathematics 9
where 119862119871MLE is given by (11) and 1205942
21198991minus120572function
which represents the lower 1 minus 120572 percentile of 12059422119899
(d) If 119862119871ge LBMLE then count = 1 else count = 0
Step 4
(a) Step 3 is repeated 1000 times(b) The estimation of confidence level (1 minus 120572) is (1 minus 120572) =
(total count)1000
Step 5
(a) Repeat Step 2ndashStep 4 for 100 times then we can getthe 100 estimations of confidence level as follows
(1 minus 120572)1 (1 minus 120572)
2 (1 minus 120572)
100 (47)
(b) The average empirical confidence level 1 minus 120572
of (1 minus 120572)119894 119894 = 1 100 that is 1 minus 120572 =
(1100)sum100
119894=1(1 minus 120572)
119894
(c) The sample mean square error (SMSE) of(1 minus 120572)
1 (1 minus 120572)
2 (1 minus 120572)
100SMSE = (1100)
times sum100
119894=1[(1 minus 120572)
119894minus (1 minus 120572)]
2
The results of simulation are summarized in Table 2 basedon 119871 = 10 the different value of sample size 119899 priorparameter (119886 119887) = (2 5) (6 15) and (2 2) at 120572 = 005respectively The scope of SMSE is between 000413 and000593
Step 1 Given 119899 119886 119887 119871 and 120572 where 119886 gt 0 119887 gt 0
Step 2 For the values of prior parameters (119886 119887) use (15) togenerate 120579 from the square-root inverted-gamma distribu-tion
Step 3
(a) The generation of data (1198851 1198852 119885
119899) is by standard
exponential distribution with parameter 1(b) Set1198842
1= 212057921198851and1198842119894= 21205792119885119894+1198842
119894minus1 for 119894 = 2 119899
1198841= radic21205792119885
1 119884
119894= radic21205792119885
119894+ 1198842
119894minus1 for 119894 = 2 119899
(48)
(1198841 1198842 119884
119899) are the upper record values from the
Rayleigh distribution(c) The values of LBBS LBBL and LBBG are calculated by
(25) (36) and (44) respectively(d) If119862
119871ge LB then count = 1 else count = 0 where LB =
LBBS or LBBL or LBBG
Step 4
(a) Step 3 is repeated 1000 times(b) The estimation of credible level (1 minus 120572) is (1 minus 120572) =
(total count)1000
Step 5
(a) Repeat Step 2ndashStep 4 for 100 times then we can getthe 100 estimations of credible level as follows
(1 minus 120572)1 (1 minus 120572)
2 (1 minus 120572)
100 (49)
(b) The average empirical credible level 1 minus 120572 of (1 minus 120572)119894
119894 = 1 100 that is 1 minus 120572 = (1100)sum100119894=1(1 minus 120572)
119894
(c) The sample mean square error (SMSE) of(1 minus 120572)
1 (1 minus 120572)
2 (1 minus 120572)
100SMSE = (1100)
times sum100
119894=1[(1 minus 120572)
119894minus (1 minus 120572)]
2
Based on 119871 = 10 the different value of sample size119899 prior parameter (119886 119887) = (2 5) (6 15) (2 2) and 120572 =
005 the results of simulation are summarized in Tables 3ndash5under SE loss function LINEX loss function with parameter119888 = minus05 05 15 and GE loss function with parameter119902 = 3 5 8 respectively The following points can be drawn
(a) All of the SMSEs are small enough and the scope ofSMSE is between 000346 and 000467
(b) Fix the size 119899 comparison of prior parameter (119886 119887) =(2 5) (6 15) and (2 2) as follows
(i) the values of SMSEwith prior parameter (119886 119887) =(6 15) and (2 5) are smaller than (119886 119887) = (2 2)respectively
(ii) a comparison of prior parameter (119886 119887) = (2 5)and (2 2) that is fix prior parameter 119886 if priorparameter 119887 increases then it can be seen thatthe SMSE will decrease
(c) In Table 4 fix the size 119899 and the prior parameter (119886 119887)comparison the parameter of LINEX loss function 119888 =minus05 05 15 as followsthe values of SMSE with prior parameter 119888 = 05 15andminus15 are the same as the values of SMSE inTable 3
(d) In Table 5 fix the prior parameter (119886 119887) comparisonof GE loss function parameter 119902 = 3 5 8 as followsthe values of SMSE with prior parameter 119902 = 3 5 and8 are the same as the values of SMSE in Table 3
Hence these results from simulation studies illustrate thatthe performance of our proposed method is acceptableMoreover we suggest that the prior parameter (119886 119887) = (6 15)and (2 5) is appropriate for the square-root inverted-gammadistribution
7 Numerical Example
In this section we propose the new hypothesis testingprocedures to a real-life data In the following examplewe discuss a real-life data for 25 ball bearings to illustratethe use of the new hypothesis testing procedures in thelifetime performance of ball bearings The proposed testingprocedures not only can handle nonnormal lifetime data butalso can handle the upper record values
10 Journal of Applied Mathematics
Table 2 Average empirical confidence level (1 minus 120572) for 119862119871under MLE when 120572 = 005
119899 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
5 093653 (000462) 093482 (000467) 093577 (000507)10 093643 (000449) 093581 (000413) 093450 (000593)15 093523 (000489) 093557 (000503) 093463 (000516)119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 3 Average empirical credible level (1 minus 120572) for 119862119871under SE loss function when 120572 = 005
119899 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
5 094357 (000386) 093760 (000386) 093925 (000395)10 094244 (000359) 093704 (000359) 093664 (000462)15 094096 (000346) 093619 (000426) 093635 (000467)119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 4 Average empirical credible level (1 minus 120572) for 119862119871under LINEX loss function when 120572 = 005
119899 119888 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
minus05 094357 (000386) 093760 (000386) 093925 (000395)5 05 094357 (000386) 093760 (000386) 093925 (000395)
15 094357 (000386) 093760 (000386) 093925 (000395)minus05 094244 (000359) 093704 (000359) 093664 (000462)
10 05 094244 (000359) 093704 (000359) 093664 (000462)15 094244 (000359) 093704 (000359) 093664 (000462)minus05 094096 (000346) 093619 (000426) 093635 (000467)
15 05 094096 (000346) 093619 (000426) 093635 (000467)15 094096 (000346) 093619 (000426) 093635 (000467)
119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 5 Average empirical credible level (1 minus 120572) for 119862119871under GE loss function when 120572 = 005
119899 119902 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
3 094357 (000386) 093760 (000386) 093925 (000395)5 5 094357 (000386) 093760 (000386) 093925 (000395)
8 094357 (000386) 093760 (000386) 093925 (000395)3 094244 (000359) 093704 (000359) 093664 (000462)
10 5 094244 (000359) 093704 (000359) 093664 (000462)8 094244 (000359) 093704 (000359) 093664 (000462)3 094096 (000346) 093619 (000426) 093635 (000467)
15 5 094096 (000346) 093619 (000426) 093635 (000467)8 094096 (000346) 093619 (000426) 093635 (000467)
119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Example The data is the failure times of 25 ball bearings inendurance testThe 25 observations are the number ofmillionrevolutions before failure for each of the ball bearings Thedata come from Caroni [27] as follows
6780 6780 6780 6864 3300 6864 9864 12804 42122892 4560 5184 5556 17340 4848 1788 9312 54124152 5196 12792 8412 10512 10584 and 6888
Raqab and Madi [28] and Lee [29] indicated that a one-parameter Rayleigh distribution is acceptable for these dataIn addition we also have a way to test the hypothesis thatthe failure data come from the Rayleigh distribution Thetesting hypothesis119867
0 119883 comes from a Rayleigh distribution
versus1198671 119883 does not come from a Rayleigh distribution is
constructed under significance level 120572 = 005From a probability plot of Figure 1 by using the Minitab
Statistical Software we can conclude the operational lifetimesdata of 25 ball bearings from the Rayleigh distribution whichis the Weibull distribution with the shape 2 (also see Lee etal [14]) For informative prior we use the prior information119864(120579) = 120579MLE ≐ 54834 and Var(120579) = 02 giving the priorparameter values as (119886 = 6014 119887 = 1001)
Here if only the upper record values have been observedthese are 119909
119880(119894) 119894 = 1 5(= 119899) = 6780 6864 9864
12804 17340
Journal of Applied Mathematics 11
10010
99
90807060504030
20
10
5
32
1
Xi
Probability plot of XiWeibull-95 CI
()
Shape 2
Scale 8002N 25
AD 0431P -value 057
Figure 1 Probability plot for failure times of 25 ball bearings data
(1) The proposed testing procedure of 119862119871with 119862
119871MLE isstated as follows
Step 1 The record values from the above data are (119909119880(119894) 119894 =
1 2 3 4 5) = (6780 6864 9864 12804 17340)
Step 2 The lower specification limit 119871 is assumed to be 2337The deal with the product managersrsquo concerns regardinglifetime performance and the conforming rate 119875
119903of products
is required to exceed 80 percent Referring to Table 1 119862119871is
required to exceed 090Thus the performance index value isset at 119888
0= 090 The testing hypothesis119867
0 119862119871le 090 versus
1198671 119862119871gt 090 is constructed
Step 3 Specify a significance level 120572 = 005
Step 4 We can calculate that the 95 lower confidenceinterval bound for 119862
119871 where
LBMLE = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871MLE)(
1205942
21198991minus120572
2119899)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 126251254)(
1205942
10095
10)
12
= 103285
(50)So the 95 one-sided confidence interval for 119862
119871is
[LBMLEinfin) = [103285infin)
Step 5 Because of the performance index value 1198880= 090 notin
[LBMLEinfin) we reject the null hypothesis1198670 119862119871 le 090
Thus we can conclude that the lifetime performanceindex of data meets the required level
(2) The proposed testing procedures of 119862119871with 119862
119871BS119862119871BL and 119862119871BG are stated as follows
Step 1 The record values from the above data are (119909119880(119894) 119894 =
1 2 3 4 5) = (6780 6864 9864 12804 17340)
Step 2 The lower specification limit 119871 is assumed to be 2337The deal with the product managersrsquo concerns regardinglifetime performance and the conforming rate 119875
119903of products
is required to exceed 80 percent Referring to Table 1 119862119871is
required to exceed 090Thus the performance index value isset at 119888
0= 090 The testing hypothesis119867
0 119862119871le 090 versus
1198671 119862119871gt 090 is constructed
Step 3 Specify a significance level 120572 = 005 the parameter ofLINEX loss function 119888 = 05 and the parameter of GE lossfunction 119902 = 2
Step 4 We can calculate the 95 lower credible intervalbound for 119862
119871 where
LBBS = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BS)
timesΓ (119899 + 119887 minus 12)
Γ (119899 + 119887)(1205942
2(119899+119887)1minus120572
2)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1246074685)
timesΓ (5 + 1001 minus 12)
Γ (5 + 1001)(1205942
2(5+1001)095
2)
12
= 096984
(51)
12 Journal of Applied Mathematics
So the 95 one-sided credible interval for 119862119871
is[LBBSinfin) = [096984infin)
LBBL = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1129561059)
times [1
2 (05)(1 minus 119890
minus05(5+1001+1)) 1205942
2(5+1001)095]
12
= 096984
(52)
So the 95 one-sided credible interval for 119862119871
is[LBBLinfin) = [096984infin)
LBBG = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times [Γ (119899 + 119887 + 1199022)
Τ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1200432998)
times [Γ (5 + 1001 + 22)
Γ (5 + 1001)]
minus12
(1205942
2(5+1001)095
2)
12
= 096984
(53)
So the 95 one-sided credible interval for 119862119871
is[LBBGinfin) = [096984infin)
Step 5 Because of the performance index value 1198880= 090 notin
[LBBSinfin) = [LBBLinfin) = [LBBGinfin) we reject the nullhypothesis119867
0 119862119871le 090
Thus we can conclude that the lifetime performance index ofdata meets the required level In addition by using the priorparameter values (119886 119887) = (6 15) based on the simulationstudies we also obtain that 119888
0= 090 notin [LBBSinfin) =
[LBBLinfin) = [LBBGinfin) = 094033 So we also rejectthe null hypothesis 119867
0 119862119871le 090 Hence the lifetime
performance index of data meets the required level
8 Conclusions
Montgomery [1] proposed a process capability index 119862119871for
larger-the-better quality characteristic The assumption ofmost process capability is normal distribution but it is ofteninvalid In this paper we consider that Rayleigh distributionis the special case of Weibull distribution
Moreover in lifetime data testing experiments the exper-imenter may not always be in a position to observe the life
times of all the products put to test In industry and reliabilitystudies many products fail under stress for example anelectronic component ceases to function in an environmentof too high temperature and a battery dies under the stressof time So in such experiments measurement may be madesequentially and only the record values are observed
In order to let the process capability indices can beeffectively used This study constructs MLE and Bayesianestimator of 119862
119871under assuming the conjugate prior distribu-
tion and SE loss function LINEX loss function and GE lossfunction based on the upper record values from the Rayleighdistribution The Bayesian estimator of 119862
119871is then utilized to
develop a credible interval in the condition of known 119871This study also provides a table of the lifetime perfor-
mance index with its corresponding conforming rate basedon the Rayleigh distributionThat is the conforming rate andthe corresponding 119862
119871can be obtained
Further the Bayesian estimator and posterior distributionof 119862119871are also utilized to construct the testing procedure of
119862119871which is based on a credible interval If you want to test
whether the products meet the required level you can utilizethe proposed testing procedure which is easily applied andan effectively method to test in the condition of known 119871Numerical example is illustrated to show that the proposedtesting procedure is effectively evaluating whether the trueperformance index meets requirements
In addition these results from simulation studies illus-trate that the performance of our proposedmethod is accept-able According to SMSE the Bayesian approach is smallerthan the non-Bayesian approach we suggest that the Bayesianapproach is better than the non-Bayesian approachThen theSMSEs of prior parameters (119886 119887) = (6 15) and (2 5) aresmaller than the SMSE of prior parameter (119886 119887) = (2 2)We suggest that the prior parameters (119886 119887) = (6 15) and(2 5) are appropriate for the square-root inverted-gammadistribution based on the Bayesian estimators under theRayleigh distribution
Acknowledgments
The authors thank the referees for their suggestions andhelpful comments on revising the paper This research waspartially supported by the National Science Council China(Plans nos NSC 101-2221-E-158-003 NSC 99-2118-M-415-001 and NSC 100-2118-M-415-002) Moreover J W Wu andC W Hong have a direct financial relation with the NationalScience Council Taiwan but other authors do not a directfinancial relation with the National Science Council Taiwan
References
[1] D C Montgomery Introduction To Statistical Quality ControlJohn Wiley amp Sons New York NY USA 1985
[2] L I Tong K S Chen and H T Chen ldquoStatistical testingfor assessing the performance of lifetime index of electroniccomponents with exponential distributionrdquo Journal of QualityReliability Management vol 19 pp 812ndash824 2002
[3] C-W Hong J-W Wu and C-H Cheng ldquoComputationalprocedure of performance assessment of lifetime index of
Journal of Applied Mathematics 13
businesses for the Pareto lifetime model with the right type IIcensored samplerdquo Applied Mathematics and Computation vol184 no 2 pp 336ndash350 2007
[4] C W Hong J W Wu and C H Cheng ldquoComputational pro-cedure of performance assessment of lifetime index of Paretolifetime businesses based on confidence intervalrdquo Applied SoftComputing vol 8 no 1 pp 698ndash705 2008
[5] J-W Wu H-M Lee and C-L Lei ldquoComputational testingalgorithmic procedure of assessment for lifetime performanceindex of products with two-parameter exponential distribu-tionrdquo Applied Mathematics and Computation vol 190 no 1 pp116ndash125 2007
[6] W-C Lee J-W Wu and C-W Hong ldquoAssessing the lifetimeperformance index of products with the exponential distribu-tion under progressively type II right censored samplesrdquo Journalof Computational and Applied Mathematics vol 231 no 2 pp648ndash656 2009
[7] W C Lee J W Wu and C L Lei ldquoEvaluating the lifetimeperformance index for the exponential lifetime productsrdquoApplied Mathematical Modelling vol 34 no 5 pp 1217ndash12242010
[8] K N Chandler ldquoThe distribution and frequency of recordvaluesrdquo Journal of the Royal Statistical Society B vol 14 pp 220ndash228 1952
[9] S K Bhattacharya and R K Tyagi ldquoBayesian survival analysisbased on the rayleigh modelrdquo Trabajos de Estadistica vol 5 no1 pp 81ndash92 1990
[10] A D Cliff and J K Ord ldquoModel building and the analysis ofspatial pattern in human geography (with discussion)rdquo Journalof the Royal Statistical Society B vol 37 no 3 pp 297ndash348 1975
[11] D D Dyer and C W Whisenand ldquoBest linear unbiasedestimator of the parameter of the Rayleigh distribution I Smallsample theory for censored order statisticsrdquo IEEE Transactionson Reliability vol R-22 no 1 pp 27ndash34 1973
[12] A A Soliman and F M Al-Aboud ldquoBayesian inference usingrecord values from Rayleigh model with applicationrdquo EuropeanJournal of Operational Research vol 185 no 2 pp 659ndash6722008
[13] StatSoft STATISTICA Software Version 7 StatSoft Tulsa Okla2005
[14] W-C Lee J-W Wu M-L Hong L-S Lin and R-L ChanldquoAssessing the lifetime performance index of Rayleigh productsbased on the Bayesian estimation under progressive type IIright censored samplesrdquo Journal of Computational and AppliedMathematics vol 235 no 6 pp 1676ndash1688 2011
[15] M Ahsanullah Record Statistics Nova Science New York NYUSA 1995
[16] B C Arnold N Balakrishnan and H N Nagaraja RecordsJohn Wiley amp Sons New York NY USA 1998
[17] P W Zehna ldquoInvariance of maximum likelihood estimatorsrdquoAnnals of Mathematical Statistics vol 37 p 744 1966
[18] A Zellner ldquoBayesian estimation and prediction using asymmet-ric loss functionsrdquo Journal of the American Statistical Associa-tion vol 81 no 394 pp 446ndash451 1986
[19] S-Y Huang ldquoEmpirical Bayes testing procedures in somenonexponential families using asymmetric linex loss functionrdquoJournal of Statistical Planning and Inference vol 46 no 3 pp293ndash309 1995
[20] R Calabria andG Pulcini ldquoPoint estimation under asymmetricloss functions for left-truncated exponential samplesrdquo Commu-nications in StatisticsmdashTheory and Methods vol 25 no 3 pp585ndash600 1996
[21] A T K Wan G Zou and A H Lee ldquoMinimax and Γ-minimaxestimation for the Poisson distribution under LINEX loss whenthe parameter space is restrictedrdquo Statistics and ProbabilityLetters vol 50 no 1 pp 23ndash32 2000
[22] S Anatolyev ldquoKernel estimation under linear-exponential lossrdquoEconomics Letters vol 91 no 1 pp 39ndash43 2006
[23] X Li Y Shi J Wei and J Chai ldquoEmpirical Bayes estimators ofreliability performances using LINEX loss under progressivelytype-II censored samplesrdquo Mathematics and Computers inSimulation vol 73 no 5 pp 320ndash326 2007
[24] D K Dyer and P S L Liu ldquoOn comparison of estimators in ageneralized life modelrdquo Microelectronics Reliability vol 32 no1-2 pp 207ndash221 1992
[25] A A Soliman ldquoEstimation of parameters of life from progres-sively censored data using Burr-XII modelrdquo IEEE Transactionson Reliability vol 54 no 1 pp 34ndash42 2005
[26] A A Soliman and G R Elkahlout ldquoBayes estimation of thelogistic distribution based on progressively censored samplesrdquoJournal of Applied Statistical Science vol 14 no 3-4 pp 281ndash2932005
[27] C Caroni ldquoThe correct ldquoball bearingsrdquo datardquo Lifetime DataAnalysis vol 8 no 4 pp 395ndash399 2002
[28] M Z Raqab and M T Madi ldquoBayesian prediction of the totaltime on test using doubly censored Rayleigh datardquo Journal ofStatistical Computation and Simulation vol 72 no 10 pp 781ndash789 2002
[29] W C Lee ldquoStatistical testing for assessing lifetime performanceIndex of the Rayleigh lifetime productsrdquo Journal of the ChineseInstitute of Industrial Engineers vol 25 pp 433ndash445 2008
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Journal of Applied Mathematics 3
index 119862119871can be rewritten as
119862119871=120583 minus 119871
120590=radic1205872120579 minus 119871
radic(4 minus 120587) 2120579= radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579
minusinfin lt 119862119871lt radic
120587
4 minus 120587
(5)
where 120583 = 119864(119883) = radic1205872120579 is the process mean 120590 =
radicVar(119883) = radic(4 minus 120587)2120579 is the process standard deviationand 119871 is the lower specification limit
3 The Conforming Rate
If the lifetime of a product 119883 exceeds the lower specificationlimit 119871 then the product is defined as a conforming productThe ratio of conforming products is known as the conformingrate and can be defined as
119875119903= 119875 (119883 ge 119871) = 119890
minus(12)(radic1205872minusradic(4minus120587)2119862119871)2
minus infin lt 119862119871lt radic120587 (4 minus 120587)
(6)
Obviously a strictly increasing relationship existsbetween the conforming rate 119875
119903and the lifetime
performance index 119862119871 Table 1 lists various 119862
119871values and
the corresponding conforming rate 119875119903by using STATISTICA
software [13] (also see [14])For the 119862
119871values which are not listed in Table 1 the
conforming rate 119875119903can be obtained through interpolation In
addition since a one-to-one mathematical relationship existsbetween the conforming rate119875
119903and the lifetime performance
index 119862119871 Therefore utilizing the one-to-one relationship
between 119875119903and 119862
119871 lifetime performance index can be
a flexible and effective tool not only evaluating productperformance but also estimating the conforming rate 119875
119903
4 Testing Procedure for the LifetimePerformance Index 119862
119871with
Non-Bayesian Approach
This section will apply non-Bayesian approach to construct amaximum likelihood estimator (MLE) of119862
119871under aRayleigh
distribution with upper record values The MLE of 119862119871is
then utilized to develop a new hypothesis testing procedurein the condition of known 119871 Assuming that the requiredindex value of lifetime performance is larger than 119888
0 where
1198880denotes the target value the null hypothesis 119867
0 119862119871le 1198880
and the alternative hypothesis1198671 119862119871gt 1198880are constructed
Based on the new hypothesis testing procedure the lifetimeperformance of products is easy to assess
Let 119883 be the lifetime of such a product and 119883 hasa Rayleigh distribution with the pdf as given by (2)Let 119883sim
= (119883119880(1) 119883119880(2) 119883
119880(119899)) be the first 119899 upper
record values arising from a sequence of iid Rayleigh
variables with pdf as given by (2) Since the joint pdf of(119883119880(1) 119883119880(2) 119883
119880(119899)) is
119891 (119909119880(119899))
119899minus1
prod
119894=1
119891 (119909119880(119894))
1 minus 119865 (119909119880(119894)) (7)
where 119891(119909) and 119865(119909) are the pdf and cdf of 119883 respec-tively (also see [12 15 16]) So the likelihood function of(119883119880(1) 119883119880(2) 119883
119880(119899)) is given as
119871 (120579 | 119909119880(1) 119909
119880(119899)) =
119899
prod
119894=1
119909119880(119894)
1205792 exp(minus
1199092
119880(119899)
21205792) (8)
The natural logarithm of the likelihood function with (8) is
ln 119871 (120579 | 119909119880(1) 119909
119880(119899)) =
119899
sum
119894=1
ln (119909119880(119894)) minus 2119899 ln 120579 minus
1199092
119880(119899)
21205792
(9)
Upon differentiating (9) with respect to 120579 and equating theresult to zero theMLE of the parameter 120579 can be shown to be
120579MLE =119883119880(119899)
radic2119899(10)
(also see [12]) By using the invariance of MLE (see Zehna[17]) the MLE of 119862
119871can be written as given by
119862119871MLE = radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579MLE
= radic120587
4 minus 120587minus radic
2
4 minus 120587sdot 119871 sdot
radic2119899
119883119880(119899)
(11)
Construct a statistical testing procedure to assess whetherthe lifetime performance index adheres to the required levelThe one-sided confidence interval for 119862
119871is obtained using
the pivotal quantity 1198832119880(119899)1205792 By using the pivotal quantity
1198832
119880(119899)1205792 given the specified significance level 120572 the level
100(1 minus 120572) one-sided confidence interval for 119862119871can be
derived as followsSince the pivotal quantity 1198832
119880(119899)1205792sim 1205942
2119899 and 1205942
21198991minus120572
function which represents the lower 1 minus 120572 percentile of 12059422119899
then
119875(1198832
119880(119899)
1205792le 1205942
21198991minus120572) = 1 minus 120572
997904rArr 119875(119862119871ge radic
120587
4 minus 120587minus(radic
120587
4 minus 120587minus119862119871MLE)(
1205942
21198991minus120572
2119899)
12
)
= 1 minus 120572
(12)
where 119862119871= radic120587(4 minus 120587) minus radic2(4 minus 120587)(119871120579) From (12) we
4 Journal of Applied Mathematics
Table 1 The lifetime performance index 119862119871versus the conforming rate 119875
119903
119862119871
119875119903
119862119871
119875119903
119862119871
119875119903
minusinfin 0000000 040 0611832 120 0896627minus450 0000147 045 0631685 125 0909965minus400 0000551 050 0651484 130 0922511minus350 0001858 055 0671182 135 0934227minus300 0005628 060 0690734 140 0945076minus250 0015308 065 0710094 145 0955027minus200 0037404 070 0729213 150 0964047minus150 0082094 075 0748044 155 0972109minus100 0161849 080 0766539 160 0979187minus050 0286621 085 0784648 165 0985259000 0455938 090 0802324 170 0990306015 0513214 095 0819518 175 0994310020 0532718 100 0836183 180 0997261025 0552370 105 0852271 185 0999147030 0572132 110 0867738 190 0999963035 0591967 115 0882538 191 0999997
obtain that a 100(1 minus 120572) one-sided confidence interval for119862119871is
119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871MLE)(
1205942
21198991minus120572
2119899)
12
(13)
where119862119871MLE is given by (11)Therefore the 100(1minus120572) lower
confidence interval bound for 119862119871can be written as
LBMLE = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871MLE)(
1205942
21198991minus120572
2119899)
12
(14)
where 119862119871MLE 120572 and 119899 denote the MLE of 119862
119871 the specified
significance level and the upper record sample of sizerespectively
The managers can use the one-sided confidence intervalto determine whether the product performance adheres tothe required levelThe proposed testing procedure of119862
119871with
119862119871MLE can be organized as follows
Step 1 Determine the lower lifetime limit 119871 for productsand the performance index value 119888
0 then the testing null
hypothesis 1198670 119862119871le 1198880and the alternative hypothesis
1198671 119862119871gt 1198880are constructed
Step 2 Specify a significance level 120572
Step 3 Given the upper record values 119909sim= (119909119880(1) 119909119880(2)
119909119880(119899)) the lower lifetime limit 119871 and the significance level 120572
then we can calculate the 100(1 minus 120572) one-sided confidenceinterval [LBMLEinfin) for 119862119871 where LBMLE as the definition of(14)
Step 4 The decision rule of statistical test is provided asfollows
(1) if the performance index value 1198880notin [LBMLEinfin) then
we will reject 1198670 It is concluded that the lifetime
performance index of products meets the requiredlevel
(2) if the performance index value 1198880isin [LBMLEinfin) then
we do not reject 1198670 It is concluded that the lifetime
performance index of products does not meet therequired level
5 Testing Procedure forthe Lifetime Performance Index 119862
119871with
the Bayesian Approach
This section will apply the Bayesian approach to constructan estimator of 119862
119871under a Rayleigh distribution with upper
record values The Bayesian estimators are developed undersymmetric and nonsymmetric loss functionsThe symmetricloss function is squared-error (SE) loss function The non-symmetric loss function includes linear exponential (LINEX)and general entropy (GE) loss functions The Bayesian esti-mator of 119862
119871is then utilized to develop a new hypothesis
testing procedure in the condition of known 119871 Assumingthat the required index value of lifetime performance is largerthan 119888
0 where 119888
0denotes the target value the null hypothesis
1198670 119862119871le 1198880and the alternative hypothesis119867
1 119862119871gt 1198880are
constructed Based on the new hypothesis testing procedurethe lifetime performance of products is easy to assess
51 Testing Procedure for the Lifetime Performance Index119862119871with the Bayesian Estimator under Squared-Error Loss
Function Let 119883 be the lifetime of such a product and 119883has a Rayleigh distribution with the pdf as given by (2)We consider the conjugate prior distribution of the formwhich is defined by the square-root inverted-gamma density
Journal of Applied Mathematics 5
as follows
1205871 (120579) =
119886119887
Γ (119887) 2119887minus1120579minus2119887minus1
119890minus1198862120579
2
(15)
where 120579 gt 0 119886 gt 0 and 119887 gt 0
119864 (120579) = (119886
2)
12Γ (119887 minus 12)
Γ (119887)
Var (120579) = 119886
2 (119887 minus 1)minus (
119886
2) [Γ (119887 minus 12)
Γ (119887)]
2
119887 gt 1
(16)
With record values 119899 products (or items) take place ontest Consider (119883
119880(1) 119883119880(2) 119883
119880(119899)) is the upper record
values We can obtain that the posterior pdf of 120579 | (119883119880(1)
119883119880(2) 119883
119880(119899)) is given as
1199011(120579 | 119909
119880(1) 119909
119880(119899))
=119871 (120579 | 119909
119880(1) 119909
119880(119899)) 1205871 (120579)
intinfin
0119871 (120579 | 119909
119880(1) 119909
119880(119899)) 1205871 (120579) 119889120579
=
120579minus2(119899+119887)minus1
(1199092
119880(119899)+ 119886)(119899+119887)
119890minus(119909119880(119899)+119886)2120579
2
Γ (119899 + 119887) 2119899+119887minus1
(17)
where 120579 gt 0 and the likelihood function 119871(120579 | 119909119880(1)
119909119880(119899)) as (8)Let 119905 = 1199092
119880(119899)+119886 then the posterior pdf can be rewritten
as
1199011(120579 | 119909
119880(1) 119909
119880(119899)) =
120579minus2(119899+119887)minus1
119905(119899+119887)
119890minus1199052120579
2
Γ (119899 + 119887) 2119899+119887minus1 (18)
and we can show that 1199051205792|119909sim
=(119909119880(1)119909119880(119899))sim 1205942
2(119899+119887)
The Bayesian estimator 120579BS of 120579 based on the squarederror (SE) loss function 119871(120579lowast 120579) = (120579lowast minus120579)2 can be derived asfollows
119864 (120579 | 119909119880(1) 119909
119880(119899))
= int
infin
0
1205791199011(120579 | 119909
119880(1) 119909
119880(119899)) 119889120579
=Γ (119899 + 119887 minus 12)
Γ (119899 + 119887)(119905
2)
12
(19)
where 119905 = 1199092119880(119899)
+ 119886 119899 + 119887 minus 12 gt 0Hence the Bayesian estimator of 120579 based on SE loss
function is given by
120579BS =Γ (119899 + 119887 minus 12)
Γ (119899 + 119887)(119879
2)
12
(20)
where 119879 = 1198832
119880(119899)+ 119886 119899 + 119887 minus 12 gt 0 119886 and 119887 are the
parameters of prior distribution with density as (15)The lifetime performance index 119862
119871of Rayleigh products
can be written as
119862119871= radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579 minusinfin lt 119862
119871lt radic
120587
4 minus 120587 (21)
By using (5) and the Bayesian estimator 120579BS as (20) 119862119871BS
based on the Bayesian estimator 120579BS of 120579 is given by
119862119871BS = radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579BS
= radic120587
4 minus 120587minus radic
2
4 minus 120587sdot 119871 sdot [
Γ (119899 + 119887 minus 12)
Γ (119899 + 119887)(119879
2)
12
]
minus1
(22)
where 119879 = 1198832
119880(119899)+ 119886 119899 + 119887 minus 12 gt 0 119886 and 119887 are the
parameters of prior distribution with density as in (15)We construct a statistical testing procedure to assess
whether the lifetime performance index adheres to therequired level The one-sided credible confidence interval for119862119871are obtained using the pivotal quantity 1199051205792|
119909sim
=(119909119880(1)119909119880(119899))
By using the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899)) given the
specified significance level 120572 the level 100(1minus120572) one-sidedcredible interval for 119862
119871can be derived as follows
Since the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899))sim 1205942
2(119899+119887)
and 12059422(119899+119887)1minus120572
function which represents the lower 1 minus 120572percentile of 1205942
2(119899+119887) then
119875(119905
1205792le 1205942
2(119899+119887)1minus120572| 119909sim) = 1 minus 120572
where 119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899))
997904rArr 119875(119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BS)
timesΓ (119899 + 119887 minus 12)
Γ (119899 + 119887)(1205942
2(119899+119887)1minus120572
2)
12
| 119909sim)
= 1 minus 120572
(23)
where 119862119871as the definition of (5)
From (23) we obtain that a 100(1minus120572)one-sided credibleinterval for 119862
119871is given by
119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BS)
timesΓ (119899 + 119887 minus 12)
Γ (119899 + 119887)(1205942
2(119899+119887)1minus120572
2)
12
(24)
where 119862119871BS is given by (22)
Therefore the 100(1 minus 120572) lower credible interval boundfor 119862119871can be written as
LBBS = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BS)
timesΓ (119899 + 119887 minus 12)
Γ (119899 + 119887)(1205942
2(119899+119887)1minus120572
2)
12
(25)
6 Journal of Applied Mathematics
where 119862119871BS is given by (22) and 119887 is a parameter of prior
distribution with density as (15)The managers can use the one-sided credible interval to
determine whether the product performance adheres to therequired level The proposed testing procedure of 119862
119871with
119862119871BS can be organized as follows
Step 1 Determine the lower lifetime limit 119871 for productsand the performance index value 119888
0 then the testing null
hypothesis 1198670 119862119871le 1198880and the alternative hypothesis
1198671 119862119871gt 1198880are constructed
Step 2 Specify a significance level 120572
Step 3 Given the parameters 119886 and 119887 of prior distributionthe upper record values 119909
sim= (119909
119880(1) 119909119880(2) 119909
119880(119899)) the
lower lifetime limit 119871 and the significance level 120572 then wecan calculate the 100(1 minus 120572) one-sided credible interval[LBBSinfin) for 119862119871 where LBBS as the definition of (25)
Step 4 The decision rule of statistical test is provided asfollows
(1) if the performance index value 1198880notin [LBBSinfin) then
we will reject 1198670 It is concluded that the lifetime
performance index of products meets the requiredlevel
(2) if the performance index value 1198880isin [LBBSinfin) then
we do not reject 1198670 It is concluded that the lifetime
performance index of products does not meet therequired level
52 Testing Procedure for the Lifetime Performance Index 119862119871
with the Bayesian Estimator under Linear Exponential LossFunction Let 119883 be the lifetime of such a product and 119883has a Rayleigh distribution with the pdf as given by (2) Weconsider the linear exponential (LINEX) loss function as (alsosee [18ndash23])
119871 (Δ) prop 119890119888Δminus 119888Δ minus 1 (26)
where Δ = (120579lowast minus 120579) 119888 = 0In this paper we suppose that
Δ1= (
120579lowast
120579)
2
minus 1 (27)
where 120579lowast is an estimator of 120579By using (26)-(27) the posterior expectation of LINEX
loss function 119871(Δ1) is
119864 [119871 (Δ1) | 119909sim] = 119890minus119888119864 [119890119888(120579lowast
120579)2 | 119909sim]
minus 119888119864 [(120579lowast
120579)
2
minus 1 | 119909sim] minus 1
(28)
where 119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899))
The value of 120579lowast that minimizes 119864[119871(Δ1) | 119909sim] denoted by
120579BL is obtained by solving the equation
119889119864 [119871 (Δ1) | 119909sim]
119889120579lowast= 119890minus119888119864[119890119888(120579lowast
120579)2 sdot2119888120579lowast
1205792| 119909sim]
minus 119888119864(2120579lowast
1205792| 119909sim) = 0
(29)
that is 120579BL is the solution to the following equation
119864[120579BL1205792119890119888(120579BL120579)2 | 119909
sim] = 119890119888119864(
120579BL1205792
| 119909sim) (30)
By using (18) and (30) we have
2120579BL (119899 + 119887) sdot119905119899+119887
(119905 minus 21198881205792
BL)119899+119887+1
= 119890119888120579BL
2 (119899 + 119887)
119905
where 119905 = 1199092119880(119899)
+ 119886
(31)
Hence the Bayesian estimator of 120579 under the LINEX lossfunction is given by
120579BL = [119879
2119888(1 minus 119890
minus119888(119899+119887+1))]
12
(32)
where 119879 = 1198832
119880(119899)+ 119886 119886 and 119887 are the parameters of prior
distribution with density as (15)By using (5) and the Bayesian estimator 120579BL as (32) 119862119871BL
based on the Bayesian estimator 120579BL of 120579 is given by
119862119871BL = radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579BL
= radic120587
4 minus 120587minus radic
2
4 minus 120587119871[119879
2119888(1 minus 119890
minus119888(119899+119887+1))]
minus12
(33)
where 119879 = 1198832
119880(119899)+ 119886 119886 and 119887 are the parameters of prior
distribution with density as (15)We construct a statistical testing procedure to assess
whether the lifetime performance index adheres to therequired level The one-sided credible confidence interval for119862119871is obtained using the pivotal quantity 1199051205792|
119909sim
=(119909119880(1)119909119880(119899))
By using the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899)) given the
specified significance level 120572 the level 100(1minus120572) one-sidedcredible interval for 119862
119871can be derived as follows
Journal of Applied Mathematics 7
Since the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899))sim 1205942
2(119899+119887)
and 12059422(119899+119887)1minus120572
function which represents the lower 1 minus 120572percentile of 1205942
2(119899+119887) then
119875(119905
1205792le 1205942
2(119899+119887)1minus120572| 119909sim) = 1 minus 120572
where 119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899))
997904rArr 119875(119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
| 119909sim)
= 1 minus 120572
(34)
where 119862119871as the definition of (5)
From (34) we obtain that a 100(1minus120572)one-sided credibleinterval for 119862
119871is
119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
(35)
where 119862119871BL is given by (33)
Therefore the 100(1 minus 120572) lower credible interval boundfor 119862119871can be written as
LBBL = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
(36)
where 119862119871BL is given by (33) and 119887 is a parameter of prior
distribution with density as (15)The managers can use the one-sided credible interval to
determine whether the product performance attains to therequired level The proposed testing procedure of 119862
119871with
119862119871BL can be organized as follows
Step 1 Determine the lower lifetime limit 119871 for productsand the performance index value 119888
0 then the testing null
hypothesis 1198670 119862119871le 1198880and the alternative hypothesis
1198671 119862119871gt 1198880are constructed
Step 2 Specify a significance level 120572
Step 3 Given the parameters 119886 and 119887 of prior distribution theparameter 119888 of LINEX loss function the upper record values119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899)) the lower lifetime limit 119871 and the
significance level 120572 then we can calculate the 100(1 minus 120572)one-sided credible interval [LBBLinfin) for 119862119871 where LBBL asthe definition of (36)
Step 4 The decision rule of statistical test is provided asfollows
(1) if the performance index value 1198880notin [LBBLinfin) then
we will reject 1198670 It is concluded that the lifetime
performance index of products meets the requiredlevel
(2) if the performance index value 1198880isin [LBBLinfin) then
we do not reject 1198670 It is concluded that the lifetime
performance index of products does not meet therequired level
53 Testing Procedure for the Lifetime Performance Index119862119871with the Bayesian Estimator under General Entropy Loss
Function Let 119883 be the lifetime of such a product and 119883has a Rayleigh distribution with the pdf as given by (2)We consider the general entropy (GE) loss function (also see[12 20])
119871 (120579lowast 120579) prop (
120579lowast
120579)
119902
minus 119902 log(120579lowast
120579) minus 1 (37)
whose minimum occurs at 120579lowast = 120579The loss function is a generalization of entropy loss used
by several authors (eg Dyer and Liu [24] Soliman [25] andSoliman and Elkahlout [26]) where the shape parameter 119902 =1 When 119902 gt 0 a positive error (120579lowast gt 120579) causes more seriousconsequences than a negative error The Bayesian estimator120579BG of 120579 under GE loss function is given by
120579BG = [119864 (120579minus119902| 119909119880(1) 119909
119880(119899))]minus1119902
(38)
By using (18) and (38) then the Bayesian estimator 120579BG of120579 is derived as follows
119864 (120579minus119902| 119909119880(1) 119909
119880(119899))
= int
infin
0
120579minus1199021199011(120579 | 119909
119880(1) 119909
119880(119899)) 119889120579
= (119905
2)
minus1199022
sdotΓ (119899 + 119887 + 1199022)
Γ (119899 + 119887)
(39)
where 119905 = 1199092119880(119899)
+ 119886 Hence the Bayesian estimator 120579BG of 120579under the GE loss function is given by
120579BG = (119879
2)
12
[Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(40)
where 119879 = 1198832
119880(119899)+ 119886 119886 and 119887 are the parameters of prior
distribution with density as (15)By using (5) and the Bayesian estimator 120579BG as (40)119862
119871BGbased on the Bayesian estimator 120579BG of 120579 is given by
119862119871BG = radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579BG
= radic120587
4 minus 120587minus radic
2
4 minus 120587sdot 119871
sdot [(119879
2)
minus1199022Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
1119902
(41)
8 Journal of Applied Mathematics
where 119879 = 1198832
119880(119899)+ 119886 119886 and 119887 are the parameters of prior
distribution with density as (15)We construct a statistical testing procedure to assess
whether the lifetime performance index adheres to therequired level The one-sided credible confidence interval for119862119871is obtained using the pivotal quantity 1199051205792|
119909sim
=(119909119880(1)119909119880(119899))
By using the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899)) given the
specified significance level 120572 the level 100(1minus120572) one-sidedcredible interval for 119862
119871can be derived as follows
Since the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899))sim 1205942
2(119899+119887)
and 12059422(119899+119887)1minus120572
function which represents the lower 1 minus 120572percentile of 1205942
2(119899+119887) then
119875(119905
1205792le 1205942
2(119899+119887)1minus120572| 119909sim) = 1 minus 120572
where 119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899))
997904rArr 119875(119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times[Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
| 119909sim)
= 1 minus 120572
(42)
where 119862119871as the definition of (5)
From (42) we obtain that a 100(1 minus 120572) one-sidedcredible interval for 119862
119871is
119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times [Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
(43)
where 119862119871BG is given by (41)
Therefore the 100(1 minus 120572) lower credible interval boundfor 119862119871can be written as
LBBG = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times [Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
(44)
where 119862119871BG is given by (41) 120572 is the specified significance
level and 119887 is a parameter of prior distribution with densityas (15)
The managers can use the one-sided credible interval todetermine whether the product performance attains to therequired level The proposed testing procedure of 119862
119871with
119862119871BG can be organized as follows
Step 1 Determine the lower lifetime limit 119871 for productsand the performance index value 119888
0 then the testing null
hypothesis 1198670 119862119871le 1198880and the alternative hypothesis
1198671 119862119871gt 1198880are constructed
Step 2 Specify a significance level 120572
Step 3 Given the parameters 119886 and 119887 of prior distributionthe parameter 119902 of GE loss function the upper record values119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899)) the lower lifetime limit 119871 and the
significance level 120572 then we can calculate the 100(1 minus 120572)one-sided credible interval [LBBGinfin) for 119862119871 where LBBG asthe definition of (44)
Step 4 The decision rule of statistical test is provided asfollows
(1) if the performance index value 1198880notin [LBBGinfin) then
we will reject 1198670 It is concluded that the lifetime
performance index of products meets the requiredlevel
(2) if the performance index value 1198880isin [LBBGinfin) then
we do not reject 1198670 It is concluded that the lifetime
performance index of products does not meet therequired level
6 The Monte Carlo Simulation Algorithm ofConfidence (or Credible) Level
In this section we will report the results of a simulationstudy for confidence (or credible) level (1 minus 120572) based on a100(1 minus 120572) one-sided confidence (or credible) interval ofthe lifetime performance index 119862
119871 We considered 120572 = 005
and then generated samples from the Rayleigh distributionwith pdf as in (2) with respect to the record values
(i) The Monte Carlo simulation algorithm of confidencelevel (1 minus 120572) for 119862
119871under MLE is given in the
following steps
Step 1 Given 119899 119886 119887 119871 and 120572 where 119886 gt 0 119887 gt 0
Step 2 For the values of prior parameters (119886 119887) use (15) togenerate 120579 from the square-root inverted-gamma distribu-tion
Step 3(a) The generation of data (119885
1 1198852 119885
119899) is by standard
exponential distribution with parameter 1(b) Set1198842
1= 212057921198851and1198842119894= 21205792119885119894+1198842
119894minus1 for 119894 = 2 119899
1198841= radic21205792119885
1 119884
119894= radic21205792119885
119894+ 1198842
119894minus1 for 119894 = 2 119899
(45)
(1198841 1198842 119884
119899) are the upper record values from the
Rayleigh distribution(c) The value of LBMLE is calculated by
LBMLE = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871MLE)(
1205942
21198991minus120572
2119899)
12
(46)
Journal of Applied Mathematics 9
where 119862119871MLE is given by (11) and 1205942
21198991minus120572function
which represents the lower 1 minus 120572 percentile of 12059422119899
(d) If 119862119871ge LBMLE then count = 1 else count = 0
Step 4
(a) Step 3 is repeated 1000 times(b) The estimation of confidence level (1 minus 120572) is (1 minus 120572) =
(total count)1000
Step 5
(a) Repeat Step 2ndashStep 4 for 100 times then we can getthe 100 estimations of confidence level as follows
(1 minus 120572)1 (1 minus 120572)
2 (1 minus 120572)
100 (47)
(b) The average empirical confidence level 1 minus 120572
of (1 minus 120572)119894 119894 = 1 100 that is 1 minus 120572 =
(1100)sum100
119894=1(1 minus 120572)
119894
(c) The sample mean square error (SMSE) of(1 minus 120572)
1 (1 minus 120572)
2 (1 minus 120572)
100SMSE = (1100)
times sum100
119894=1[(1 minus 120572)
119894minus (1 minus 120572)]
2
The results of simulation are summarized in Table 2 basedon 119871 = 10 the different value of sample size 119899 priorparameter (119886 119887) = (2 5) (6 15) and (2 2) at 120572 = 005respectively The scope of SMSE is between 000413 and000593
Step 1 Given 119899 119886 119887 119871 and 120572 where 119886 gt 0 119887 gt 0
Step 2 For the values of prior parameters (119886 119887) use (15) togenerate 120579 from the square-root inverted-gamma distribu-tion
Step 3
(a) The generation of data (1198851 1198852 119885
119899) is by standard
exponential distribution with parameter 1(b) Set1198842
1= 212057921198851and1198842119894= 21205792119885119894+1198842
119894minus1 for 119894 = 2 119899
1198841= radic21205792119885
1 119884
119894= radic21205792119885
119894+ 1198842
119894minus1 for 119894 = 2 119899
(48)
(1198841 1198842 119884
119899) are the upper record values from the
Rayleigh distribution(c) The values of LBBS LBBL and LBBG are calculated by
(25) (36) and (44) respectively(d) If119862
119871ge LB then count = 1 else count = 0 where LB =
LBBS or LBBL or LBBG
Step 4
(a) Step 3 is repeated 1000 times(b) The estimation of credible level (1 minus 120572) is (1 minus 120572) =
(total count)1000
Step 5
(a) Repeat Step 2ndashStep 4 for 100 times then we can getthe 100 estimations of credible level as follows
(1 minus 120572)1 (1 minus 120572)
2 (1 minus 120572)
100 (49)
(b) The average empirical credible level 1 minus 120572 of (1 minus 120572)119894
119894 = 1 100 that is 1 minus 120572 = (1100)sum100119894=1(1 minus 120572)
119894
(c) The sample mean square error (SMSE) of(1 minus 120572)
1 (1 minus 120572)
2 (1 minus 120572)
100SMSE = (1100)
times sum100
119894=1[(1 minus 120572)
119894minus (1 minus 120572)]
2
Based on 119871 = 10 the different value of sample size119899 prior parameter (119886 119887) = (2 5) (6 15) (2 2) and 120572 =
005 the results of simulation are summarized in Tables 3ndash5under SE loss function LINEX loss function with parameter119888 = minus05 05 15 and GE loss function with parameter119902 = 3 5 8 respectively The following points can be drawn
(a) All of the SMSEs are small enough and the scope ofSMSE is between 000346 and 000467
(b) Fix the size 119899 comparison of prior parameter (119886 119887) =(2 5) (6 15) and (2 2) as follows
(i) the values of SMSEwith prior parameter (119886 119887) =(6 15) and (2 5) are smaller than (119886 119887) = (2 2)respectively
(ii) a comparison of prior parameter (119886 119887) = (2 5)and (2 2) that is fix prior parameter 119886 if priorparameter 119887 increases then it can be seen thatthe SMSE will decrease
(c) In Table 4 fix the size 119899 and the prior parameter (119886 119887)comparison the parameter of LINEX loss function 119888 =minus05 05 15 as followsthe values of SMSE with prior parameter 119888 = 05 15andminus15 are the same as the values of SMSE inTable 3
(d) In Table 5 fix the prior parameter (119886 119887) comparisonof GE loss function parameter 119902 = 3 5 8 as followsthe values of SMSE with prior parameter 119902 = 3 5 and8 are the same as the values of SMSE in Table 3
Hence these results from simulation studies illustrate thatthe performance of our proposed method is acceptableMoreover we suggest that the prior parameter (119886 119887) = (6 15)and (2 5) is appropriate for the square-root inverted-gammadistribution
7 Numerical Example
In this section we propose the new hypothesis testingprocedures to a real-life data In the following examplewe discuss a real-life data for 25 ball bearings to illustratethe use of the new hypothesis testing procedures in thelifetime performance of ball bearings The proposed testingprocedures not only can handle nonnormal lifetime data butalso can handle the upper record values
10 Journal of Applied Mathematics
Table 2 Average empirical confidence level (1 minus 120572) for 119862119871under MLE when 120572 = 005
119899 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
5 093653 (000462) 093482 (000467) 093577 (000507)10 093643 (000449) 093581 (000413) 093450 (000593)15 093523 (000489) 093557 (000503) 093463 (000516)119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 3 Average empirical credible level (1 minus 120572) for 119862119871under SE loss function when 120572 = 005
119899 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
5 094357 (000386) 093760 (000386) 093925 (000395)10 094244 (000359) 093704 (000359) 093664 (000462)15 094096 (000346) 093619 (000426) 093635 (000467)119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 4 Average empirical credible level (1 minus 120572) for 119862119871under LINEX loss function when 120572 = 005
119899 119888 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
minus05 094357 (000386) 093760 (000386) 093925 (000395)5 05 094357 (000386) 093760 (000386) 093925 (000395)
15 094357 (000386) 093760 (000386) 093925 (000395)minus05 094244 (000359) 093704 (000359) 093664 (000462)
10 05 094244 (000359) 093704 (000359) 093664 (000462)15 094244 (000359) 093704 (000359) 093664 (000462)minus05 094096 (000346) 093619 (000426) 093635 (000467)
15 05 094096 (000346) 093619 (000426) 093635 (000467)15 094096 (000346) 093619 (000426) 093635 (000467)
119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 5 Average empirical credible level (1 minus 120572) for 119862119871under GE loss function when 120572 = 005
119899 119902 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
3 094357 (000386) 093760 (000386) 093925 (000395)5 5 094357 (000386) 093760 (000386) 093925 (000395)
8 094357 (000386) 093760 (000386) 093925 (000395)3 094244 (000359) 093704 (000359) 093664 (000462)
10 5 094244 (000359) 093704 (000359) 093664 (000462)8 094244 (000359) 093704 (000359) 093664 (000462)3 094096 (000346) 093619 (000426) 093635 (000467)
15 5 094096 (000346) 093619 (000426) 093635 (000467)8 094096 (000346) 093619 (000426) 093635 (000467)
119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Example The data is the failure times of 25 ball bearings inendurance testThe 25 observations are the number ofmillionrevolutions before failure for each of the ball bearings Thedata come from Caroni [27] as follows
6780 6780 6780 6864 3300 6864 9864 12804 42122892 4560 5184 5556 17340 4848 1788 9312 54124152 5196 12792 8412 10512 10584 and 6888
Raqab and Madi [28] and Lee [29] indicated that a one-parameter Rayleigh distribution is acceptable for these dataIn addition we also have a way to test the hypothesis thatthe failure data come from the Rayleigh distribution Thetesting hypothesis119867
0 119883 comes from a Rayleigh distribution
versus1198671 119883 does not come from a Rayleigh distribution is
constructed under significance level 120572 = 005From a probability plot of Figure 1 by using the Minitab
Statistical Software we can conclude the operational lifetimesdata of 25 ball bearings from the Rayleigh distribution whichis the Weibull distribution with the shape 2 (also see Lee etal [14]) For informative prior we use the prior information119864(120579) = 120579MLE ≐ 54834 and Var(120579) = 02 giving the priorparameter values as (119886 = 6014 119887 = 1001)
Here if only the upper record values have been observedthese are 119909
119880(119894) 119894 = 1 5(= 119899) = 6780 6864 9864
12804 17340
Journal of Applied Mathematics 11
10010
99
90807060504030
20
10
5
32
1
Xi
Probability plot of XiWeibull-95 CI
()
Shape 2
Scale 8002N 25
AD 0431P -value 057
Figure 1 Probability plot for failure times of 25 ball bearings data
(1) The proposed testing procedure of 119862119871with 119862
119871MLE isstated as follows
Step 1 The record values from the above data are (119909119880(119894) 119894 =
1 2 3 4 5) = (6780 6864 9864 12804 17340)
Step 2 The lower specification limit 119871 is assumed to be 2337The deal with the product managersrsquo concerns regardinglifetime performance and the conforming rate 119875
119903of products
is required to exceed 80 percent Referring to Table 1 119862119871is
required to exceed 090Thus the performance index value isset at 119888
0= 090 The testing hypothesis119867
0 119862119871le 090 versus
1198671 119862119871gt 090 is constructed
Step 3 Specify a significance level 120572 = 005
Step 4 We can calculate that the 95 lower confidenceinterval bound for 119862
119871 where
LBMLE = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871MLE)(
1205942
21198991minus120572
2119899)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 126251254)(
1205942
10095
10)
12
= 103285
(50)So the 95 one-sided confidence interval for 119862
119871is
[LBMLEinfin) = [103285infin)
Step 5 Because of the performance index value 1198880= 090 notin
[LBMLEinfin) we reject the null hypothesis1198670 119862119871 le 090
Thus we can conclude that the lifetime performanceindex of data meets the required level
(2) The proposed testing procedures of 119862119871with 119862
119871BS119862119871BL and 119862119871BG are stated as follows
Step 1 The record values from the above data are (119909119880(119894) 119894 =
1 2 3 4 5) = (6780 6864 9864 12804 17340)
Step 2 The lower specification limit 119871 is assumed to be 2337The deal with the product managersrsquo concerns regardinglifetime performance and the conforming rate 119875
119903of products
is required to exceed 80 percent Referring to Table 1 119862119871is
required to exceed 090Thus the performance index value isset at 119888
0= 090 The testing hypothesis119867
0 119862119871le 090 versus
1198671 119862119871gt 090 is constructed
Step 3 Specify a significance level 120572 = 005 the parameter ofLINEX loss function 119888 = 05 and the parameter of GE lossfunction 119902 = 2
Step 4 We can calculate the 95 lower credible intervalbound for 119862
119871 where
LBBS = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BS)
timesΓ (119899 + 119887 minus 12)
Γ (119899 + 119887)(1205942
2(119899+119887)1minus120572
2)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1246074685)
timesΓ (5 + 1001 minus 12)
Γ (5 + 1001)(1205942
2(5+1001)095
2)
12
= 096984
(51)
12 Journal of Applied Mathematics
So the 95 one-sided credible interval for 119862119871
is[LBBSinfin) = [096984infin)
LBBL = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1129561059)
times [1
2 (05)(1 minus 119890
minus05(5+1001+1)) 1205942
2(5+1001)095]
12
= 096984
(52)
So the 95 one-sided credible interval for 119862119871
is[LBBLinfin) = [096984infin)
LBBG = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times [Γ (119899 + 119887 + 1199022)
Τ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1200432998)
times [Γ (5 + 1001 + 22)
Γ (5 + 1001)]
minus12
(1205942
2(5+1001)095
2)
12
= 096984
(53)
So the 95 one-sided credible interval for 119862119871
is[LBBGinfin) = [096984infin)
Step 5 Because of the performance index value 1198880= 090 notin
[LBBSinfin) = [LBBLinfin) = [LBBGinfin) we reject the nullhypothesis119867
0 119862119871le 090
Thus we can conclude that the lifetime performance index ofdata meets the required level In addition by using the priorparameter values (119886 119887) = (6 15) based on the simulationstudies we also obtain that 119888
0= 090 notin [LBBSinfin) =
[LBBLinfin) = [LBBGinfin) = 094033 So we also rejectthe null hypothesis 119867
0 119862119871le 090 Hence the lifetime
performance index of data meets the required level
8 Conclusions
Montgomery [1] proposed a process capability index 119862119871for
larger-the-better quality characteristic The assumption ofmost process capability is normal distribution but it is ofteninvalid In this paper we consider that Rayleigh distributionis the special case of Weibull distribution
Moreover in lifetime data testing experiments the exper-imenter may not always be in a position to observe the life
times of all the products put to test In industry and reliabilitystudies many products fail under stress for example anelectronic component ceases to function in an environmentof too high temperature and a battery dies under the stressof time So in such experiments measurement may be madesequentially and only the record values are observed
In order to let the process capability indices can beeffectively used This study constructs MLE and Bayesianestimator of 119862
119871under assuming the conjugate prior distribu-
tion and SE loss function LINEX loss function and GE lossfunction based on the upper record values from the Rayleighdistribution The Bayesian estimator of 119862
119871is then utilized to
develop a credible interval in the condition of known 119871This study also provides a table of the lifetime perfor-
mance index with its corresponding conforming rate basedon the Rayleigh distributionThat is the conforming rate andthe corresponding 119862
119871can be obtained
Further the Bayesian estimator and posterior distributionof 119862119871are also utilized to construct the testing procedure of
119862119871which is based on a credible interval If you want to test
whether the products meet the required level you can utilizethe proposed testing procedure which is easily applied andan effectively method to test in the condition of known 119871Numerical example is illustrated to show that the proposedtesting procedure is effectively evaluating whether the trueperformance index meets requirements
In addition these results from simulation studies illus-trate that the performance of our proposedmethod is accept-able According to SMSE the Bayesian approach is smallerthan the non-Bayesian approach we suggest that the Bayesianapproach is better than the non-Bayesian approachThen theSMSEs of prior parameters (119886 119887) = (6 15) and (2 5) aresmaller than the SMSE of prior parameter (119886 119887) = (2 2)We suggest that the prior parameters (119886 119887) = (6 15) and(2 5) are appropriate for the square-root inverted-gammadistribution based on the Bayesian estimators under theRayleigh distribution
Acknowledgments
The authors thank the referees for their suggestions andhelpful comments on revising the paper This research waspartially supported by the National Science Council China(Plans nos NSC 101-2221-E-158-003 NSC 99-2118-M-415-001 and NSC 100-2118-M-415-002) Moreover J W Wu andC W Hong have a direct financial relation with the NationalScience Council Taiwan but other authors do not a directfinancial relation with the National Science Council Taiwan
References
[1] D C Montgomery Introduction To Statistical Quality ControlJohn Wiley amp Sons New York NY USA 1985
[2] L I Tong K S Chen and H T Chen ldquoStatistical testingfor assessing the performance of lifetime index of electroniccomponents with exponential distributionrdquo Journal of QualityReliability Management vol 19 pp 812ndash824 2002
[3] C-W Hong J-W Wu and C-H Cheng ldquoComputationalprocedure of performance assessment of lifetime index of
Journal of Applied Mathematics 13
businesses for the Pareto lifetime model with the right type IIcensored samplerdquo Applied Mathematics and Computation vol184 no 2 pp 336ndash350 2007
[4] C W Hong J W Wu and C H Cheng ldquoComputational pro-cedure of performance assessment of lifetime index of Paretolifetime businesses based on confidence intervalrdquo Applied SoftComputing vol 8 no 1 pp 698ndash705 2008
[5] J-W Wu H-M Lee and C-L Lei ldquoComputational testingalgorithmic procedure of assessment for lifetime performanceindex of products with two-parameter exponential distribu-tionrdquo Applied Mathematics and Computation vol 190 no 1 pp116ndash125 2007
[6] W-C Lee J-W Wu and C-W Hong ldquoAssessing the lifetimeperformance index of products with the exponential distribu-tion under progressively type II right censored samplesrdquo Journalof Computational and Applied Mathematics vol 231 no 2 pp648ndash656 2009
[7] W C Lee J W Wu and C L Lei ldquoEvaluating the lifetimeperformance index for the exponential lifetime productsrdquoApplied Mathematical Modelling vol 34 no 5 pp 1217ndash12242010
[8] K N Chandler ldquoThe distribution and frequency of recordvaluesrdquo Journal of the Royal Statistical Society B vol 14 pp 220ndash228 1952
[9] S K Bhattacharya and R K Tyagi ldquoBayesian survival analysisbased on the rayleigh modelrdquo Trabajos de Estadistica vol 5 no1 pp 81ndash92 1990
[10] A D Cliff and J K Ord ldquoModel building and the analysis ofspatial pattern in human geography (with discussion)rdquo Journalof the Royal Statistical Society B vol 37 no 3 pp 297ndash348 1975
[11] D D Dyer and C W Whisenand ldquoBest linear unbiasedestimator of the parameter of the Rayleigh distribution I Smallsample theory for censored order statisticsrdquo IEEE Transactionson Reliability vol R-22 no 1 pp 27ndash34 1973
[12] A A Soliman and F M Al-Aboud ldquoBayesian inference usingrecord values from Rayleigh model with applicationrdquo EuropeanJournal of Operational Research vol 185 no 2 pp 659ndash6722008
[13] StatSoft STATISTICA Software Version 7 StatSoft Tulsa Okla2005
[14] W-C Lee J-W Wu M-L Hong L-S Lin and R-L ChanldquoAssessing the lifetime performance index of Rayleigh productsbased on the Bayesian estimation under progressive type IIright censored samplesrdquo Journal of Computational and AppliedMathematics vol 235 no 6 pp 1676ndash1688 2011
[15] M Ahsanullah Record Statistics Nova Science New York NYUSA 1995
[16] B C Arnold N Balakrishnan and H N Nagaraja RecordsJohn Wiley amp Sons New York NY USA 1998
[17] P W Zehna ldquoInvariance of maximum likelihood estimatorsrdquoAnnals of Mathematical Statistics vol 37 p 744 1966
[18] A Zellner ldquoBayesian estimation and prediction using asymmet-ric loss functionsrdquo Journal of the American Statistical Associa-tion vol 81 no 394 pp 446ndash451 1986
[19] S-Y Huang ldquoEmpirical Bayes testing procedures in somenonexponential families using asymmetric linex loss functionrdquoJournal of Statistical Planning and Inference vol 46 no 3 pp293ndash309 1995
[20] R Calabria andG Pulcini ldquoPoint estimation under asymmetricloss functions for left-truncated exponential samplesrdquo Commu-nications in StatisticsmdashTheory and Methods vol 25 no 3 pp585ndash600 1996
[21] A T K Wan G Zou and A H Lee ldquoMinimax and Γ-minimaxestimation for the Poisson distribution under LINEX loss whenthe parameter space is restrictedrdquo Statistics and ProbabilityLetters vol 50 no 1 pp 23ndash32 2000
[22] S Anatolyev ldquoKernel estimation under linear-exponential lossrdquoEconomics Letters vol 91 no 1 pp 39ndash43 2006
[23] X Li Y Shi J Wei and J Chai ldquoEmpirical Bayes estimators ofreliability performances using LINEX loss under progressivelytype-II censored samplesrdquo Mathematics and Computers inSimulation vol 73 no 5 pp 320ndash326 2007
[24] D K Dyer and P S L Liu ldquoOn comparison of estimators in ageneralized life modelrdquo Microelectronics Reliability vol 32 no1-2 pp 207ndash221 1992
[25] A A Soliman ldquoEstimation of parameters of life from progres-sively censored data using Burr-XII modelrdquo IEEE Transactionson Reliability vol 54 no 1 pp 34ndash42 2005
[26] A A Soliman and G R Elkahlout ldquoBayes estimation of thelogistic distribution based on progressively censored samplesrdquoJournal of Applied Statistical Science vol 14 no 3-4 pp 281ndash2932005
[27] C Caroni ldquoThe correct ldquoball bearingsrdquo datardquo Lifetime DataAnalysis vol 8 no 4 pp 395ndash399 2002
[28] M Z Raqab and M T Madi ldquoBayesian prediction of the totaltime on test using doubly censored Rayleigh datardquo Journal ofStatistical Computation and Simulation vol 72 no 10 pp 781ndash789 2002
[29] W C Lee ldquoStatistical testing for assessing lifetime performanceIndex of the Rayleigh lifetime productsrdquo Journal of the ChineseInstitute of Industrial Engineers vol 25 pp 433ndash445 2008
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
4 Journal of Applied Mathematics
Table 1 The lifetime performance index 119862119871versus the conforming rate 119875
119903
119862119871
119875119903
119862119871
119875119903
119862119871
119875119903
minusinfin 0000000 040 0611832 120 0896627minus450 0000147 045 0631685 125 0909965minus400 0000551 050 0651484 130 0922511minus350 0001858 055 0671182 135 0934227minus300 0005628 060 0690734 140 0945076minus250 0015308 065 0710094 145 0955027minus200 0037404 070 0729213 150 0964047minus150 0082094 075 0748044 155 0972109minus100 0161849 080 0766539 160 0979187minus050 0286621 085 0784648 165 0985259000 0455938 090 0802324 170 0990306015 0513214 095 0819518 175 0994310020 0532718 100 0836183 180 0997261025 0552370 105 0852271 185 0999147030 0572132 110 0867738 190 0999963035 0591967 115 0882538 191 0999997
obtain that a 100(1 minus 120572) one-sided confidence interval for119862119871is
119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871MLE)(
1205942
21198991minus120572
2119899)
12
(13)
where119862119871MLE is given by (11)Therefore the 100(1minus120572) lower
confidence interval bound for 119862119871can be written as
LBMLE = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871MLE)(
1205942
21198991minus120572
2119899)
12
(14)
where 119862119871MLE 120572 and 119899 denote the MLE of 119862
119871 the specified
significance level and the upper record sample of sizerespectively
The managers can use the one-sided confidence intervalto determine whether the product performance adheres tothe required levelThe proposed testing procedure of119862
119871with
119862119871MLE can be organized as follows
Step 1 Determine the lower lifetime limit 119871 for productsand the performance index value 119888
0 then the testing null
hypothesis 1198670 119862119871le 1198880and the alternative hypothesis
1198671 119862119871gt 1198880are constructed
Step 2 Specify a significance level 120572
Step 3 Given the upper record values 119909sim= (119909119880(1) 119909119880(2)
119909119880(119899)) the lower lifetime limit 119871 and the significance level 120572
then we can calculate the 100(1 minus 120572) one-sided confidenceinterval [LBMLEinfin) for 119862119871 where LBMLE as the definition of(14)
Step 4 The decision rule of statistical test is provided asfollows
(1) if the performance index value 1198880notin [LBMLEinfin) then
we will reject 1198670 It is concluded that the lifetime
performance index of products meets the requiredlevel
(2) if the performance index value 1198880isin [LBMLEinfin) then
we do not reject 1198670 It is concluded that the lifetime
performance index of products does not meet therequired level
5 Testing Procedure forthe Lifetime Performance Index 119862
119871with
the Bayesian Approach
This section will apply the Bayesian approach to constructan estimator of 119862
119871under a Rayleigh distribution with upper
record values The Bayesian estimators are developed undersymmetric and nonsymmetric loss functionsThe symmetricloss function is squared-error (SE) loss function The non-symmetric loss function includes linear exponential (LINEX)and general entropy (GE) loss functions The Bayesian esti-mator of 119862
119871is then utilized to develop a new hypothesis
testing procedure in the condition of known 119871 Assumingthat the required index value of lifetime performance is largerthan 119888
0 where 119888
0denotes the target value the null hypothesis
1198670 119862119871le 1198880and the alternative hypothesis119867
1 119862119871gt 1198880are
constructed Based on the new hypothesis testing procedurethe lifetime performance of products is easy to assess
51 Testing Procedure for the Lifetime Performance Index119862119871with the Bayesian Estimator under Squared-Error Loss
Function Let 119883 be the lifetime of such a product and 119883has a Rayleigh distribution with the pdf as given by (2)We consider the conjugate prior distribution of the formwhich is defined by the square-root inverted-gamma density
Journal of Applied Mathematics 5
as follows
1205871 (120579) =
119886119887
Γ (119887) 2119887minus1120579minus2119887minus1
119890minus1198862120579
2
(15)
where 120579 gt 0 119886 gt 0 and 119887 gt 0
119864 (120579) = (119886
2)
12Γ (119887 minus 12)
Γ (119887)
Var (120579) = 119886
2 (119887 minus 1)minus (
119886
2) [Γ (119887 minus 12)
Γ (119887)]
2
119887 gt 1
(16)
With record values 119899 products (or items) take place ontest Consider (119883
119880(1) 119883119880(2) 119883
119880(119899)) is the upper record
values We can obtain that the posterior pdf of 120579 | (119883119880(1)
119883119880(2) 119883
119880(119899)) is given as
1199011(120579 | 119909
119880(1) 119909
119880(119899))
=119871 (120579 | 119909
119880(1) 119909
119880(119899)) 1205871 (120579)
intinfin
0119871 (120579 | 119909
119880(1) 119909
119880(119899)) 1205871 (120579) 119889120579
=
120579minus2(119899+119887)minus1
(1199092
119880(119899)+ 119886)(119899+119887)
119890minus(119909119880(119899)+119886)2120579
2
Γ (119899 + 119887) 2119899+119887minus1
(17)
where 120579 gt 0 and the likelihood function 119871(120579 | 119909119880(1)
119909119880(119899)) as (8)Let 119905 = 1199092
119880(119899)+119886 then the posterior pdf can be rewritten
as
1199011(120579 | 119909
119880(1) 119909
119880(119899)) =
120579minus2(119899+119887)minus1
119905(119899+119887)
119890minus1199052120579
2
Γ (119899 + 119887) 2119899+119887minus1 (18)
and we can show that 1199051205792|119909sim
=(119909119880(1)119909119880(119899))sim 1205942
2(119899+119887)
The Bayesian estimator 120579BS of 120579 based on the squarederror (SE) loss function 119871(120579lowast 120579) = (120579lowast minus120579)2 can be derived asfollows
119864 (120579 | 119909119880(1) 119909
119880(119899))
= int
infin
0
1205791199011(120579 | 119909
119880(1) 119909
119880(119899)) 119889120579
=Γ (119899 + 119887 minus 12)
Γ (119899 + 119887)(119905
2)
12
(19)
where 119905 = 1199092119880(119899)
+ 119886 119899 + 119887 minus 12 gt 0Hence the Bayesian estimator of 120579 based on SE loss
function is given by
120579BS =Γ (119899 + 119887 minus 12)
Γ (119899 + 119887)(119879
2)
12
(20)
where 119879 = 1198832
119880(119899)+ 119886 119899 + 119887 minus 12 gt 0 119886 and 119887 are the
parameters of prior distribution with density as (15)The lifetime performance index 119862
119871of Rayleigh products
can be written as
119862119871= radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579 minusinfin lt 119862
119871lt radic
120587
4 minus 120587 (21)
By using (5) and the Bayesian estimator 120579BS as (20) 119862119871BS
based on the Bayesian estimator 120579BS of 120579 is given by
119862119871BS = radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579BS
= radic120587
4 minus 120587minus radic
2
4 minus 120587sdot 119871 sdot [
Γ (119899 + 119887 minus 12)
Γ (119899 + 119887)(119879
2)
12
]
minus1
(22)
where 119879 = 1198832
119880(119899)+ 119886 119899 + 119887 minus 12 gt 0 119886 and 119887 are the
parameters of prior distribution with density as in (15)We construct a statistical testing procedure to assess
whether the lifetime performance index adheres to therequired level The one-sided credible confidence interval for119862119871are obtained using the pivotal quantity 1199051205792|
119909sim
=(119909119880(1)119909119880(119899))
By using the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899)) given the
specified significance level 120572 the level 100(1minus120572) one-sidedcredible interval for 119862
119871can be derived as follows
Since the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899))sim 1205942
2(119899+119887)
and 12059422(119899+119887)1minus120572
function which represents the lower 1 minus 120572percentile of 1205942
2(119899+119887) then
119875(119905
1205792le 1205942
2(119899+119887)1minus120572| 119909sim) = 1 minus 120572
where 119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899))
997904rArr 119875(119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BS)
timesΓ (119899 + 119887 minus 12)
Γ (119899 + 119887)(1205942
2(119899+119887)1minus120572
2)
12
| 119909sim)
= 1 minus 120572
(23)
where 119862119871as the definition of (5)
From (23) we obtain that a 100(1minus120572)one-sided credibleinterval for 119862
119871is given by
119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BS)
timesΓ (119899 + 119887 minus 12)
Γ (119899 + 119887)(1205942
2(119899+119887)1minus120572
2)
12
(24)
where 119862119871BS is given by (22)
Therefore the 100(1 minus 120572) lower credible interval boundfor 119862119871can be written as
LBBS = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BS)
timesΓ (119899 + 119887 minus 12)
Γ (119899 + 119887)(1205942
2(119899+119887)1minus120572
2)
12
(25)
6 Journal of Applied Mathematics
where 119862119871BS is given by (22) and 119887 is a parameter of prior
distribution with density as (15)The managers can use the one-sided credible interval to
determine whether the product performance adheres to therequired level The proposed testing procedure of 119862
119871with
119862119871BS can be organized as follows
Step 1 Determine the lower lifetime limit 119871 for productsand the performance index value 119888
0 then the testing null
hypothesis 1198670 119862119871le 1198880and the alternative hypothesis
1198671 119862119871gt 1198880are constructed
Step 2 Specify a significance level 120572
Step 3 Given the parameters 119886 and 119887 of prior distributionthe upper record values 119909
sim= (119909
119880(1) 119909119880(2) 119909
119880(119899)) the
lower lifetime limit 119871 and the significance level 120572 then wecan calculate the 100(1 minus 120572) one-sided credible interval[LBBSinfin) for 119862119871 where LBBS as the definition of (25)
Step 4 The decision rule of statistical test is provided asfollows
(1) if the performance index value 1198880notin [LBBSinfin) then
we will reject 1198670 It is concluded that the lifetime
performance index of products meets the requiredlevel
(2) if the performance index value 1198880isin [LBBSinfin) then
we do not reject 1198670 It is concluded that the lifetime
performance index of products does not meet therequired level
52 Testing Procedure for the Lifetime Performance Index 119862119871
with the Bayesian Estimator under Linear Exponential LossFunction Let 119883 be the lifetime of such a product and 119883has a Rayleigh distribution with the pdf as given by (2) Weconsider the linear exponential (LINEX) loss function as (alsosee [18ndash23])
119871 (Δ) prop 119890119888Δminus 119888Δ minus 1 (26)
where Δ = (120579lowast minus 120579) 119888 = 0In this paper we suppose that
Δ1= (
120579lowast
120579)
2
minus 1 (27)
where 120579lowast is an estimator of 120579By using (26)-(27) the posterior expectation of LINEX
loss function 119871(Δ1) is
119864 [119871 (Δ1) | 119909sim] = 119890minus119888119864 [119890119888(120579lowast
120579)2 | 119909sim]
minus 119888119864 [(120579lowast
120579)
2
minus 1 | 119909sim] minus 1
(28)
where 119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899))
The value of 120579lowast that minimizes 119864[119871(Δ1) | 119909sim] denoted by
120579BL is obtained by solving the equation
119889119864 [119871 (Δ1) | 119909sim]
119889120579lowast= 119890minus119888119864[119890119888(120579lowast
120579)2 sdot2119888120579lowast
1205792| 119909sim]
minus 119888119864(2120579lowast
1205792| 119909sim) = 0
(29)
that is 120579BL is the solution to the following equation
119864[120579BL1205792119890119888(120579BL120579)2 | 119909
sim] = 119890119888119864(
120579BL1205792
| 119909sim) (30)
By using (18) and (30) we have
2120579BL (119899 + 119887) sdot119905119899+119887
(119905 minus 21198881205792
BL)119899+119887+1
= 119890119888120579BL
2 (119899 + 119887)
119905
where 119905 = 1199092119880(119899)
+ 119886
(31)
Hence the Bayesian estimator of 120579 under the LINEX lossfunction is given by
120579BL = [119879
2119888(1 minus 119890
minus119888(119899+119887+1))]
12
(32)
where 119879 = 1198832
119880(119899)+ 119886 119886 and 119887 are the parameters of prior
distribution with density as (15)By using (5) and the Bayesian estimator 120579BL as (32) 119862119871BL
based on the Bayesian estimator 120579BL of 120579 is given by
119862119871BL = radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579BL
= radic120587
4 minus 120587minus radic
2
4 minus 120587119871[119879
2119888(1 minus 119890
minus119888(119899+119887+1))]
minus12
(33)
where 119879 = 1198832
119880(119899)+ 119886 119886 and 119887 are the parameters of prior
distribution with density as (15)We construct a statistical testing procedure to assess
whether the lifetime performance index adheres to therequired level The one-sided credible confidence interval for119862119871is obtained using the pivotal quantity 1199051205792|
119909sim
=(119909119880(1)119909119880(119899))
By using the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899)) given the
specified significance level 120572 the level 100(1minus120572) one-sidedcredible interval for 119862
119871can be derived as follows
Journal of Applied Mathematics 7
Since the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899))sim 1205942
2(119899+119887)
and 12059422(119899+119887)1minus120572
function which represents the lower 1 minus 120572percentile of 1205942
2(119899+119887) then
119875(119905
1205792le 1205942
2(119899+119887)1minus120572| 119909sim) = 1 minus 120572
where 119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899))
997904rArr 119875(119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
| 119909sim)
= 1 minus 120572
(34)
where 119862119871as the definition of (5)
From (34) we obtain that a 100(1minus120572)one-sided credibleinterval for 119862
119871is
119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
(35)
where 119862119871BL is given by (33)
Therefore the 100(1 minus 120572) lower credible interval boundfor 119862119871can be written as
LBBL = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
(36)
where 119862119871BL is given by (33) and 119887 is a parameter of prior
distribution with density as (15)The managers can use the one-sided credible interval to
determine whether the product performance attains to therequired level The proposed testing procedure of 119862
119871with
119862119871BL can be organized as follows
Step 1 Determine the lower lifetime limit 119871 for productsand the performance index value 119888
0 then the testing null
hypothesis 1198670 119862119871le 1198880and the alternative hypothesis
1198671 119862119871gt 1198880are constructed
Step 2 Specify a significance level 120572
Step 3 Given the parameters 119886 and 119887 of prior distribution theparameter 119888 of LINEX loss function the upper record values119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899)) the lower lifetime limit 119871 and the
significance level 120572 then we can calculate the 100(1 minus 120572)one-sided credible interval [LBBLinfin) for 119862119871 where LBBL asthe definition of (36)
Step 4 The decision rule of statistical test is provided asfollows
(1) if the performance index value 1198880notin [LBBLinfin) then
we will reject 1198670 It is concluded that the lifetime
performance index of products meets the requiredlevel
(2) if the performance index value 1198880isin [LBBLinfin) then
we do not reject 1198670 It is concluded that the lifetime
performance index of products does not meet therequired level
53 Testing Procedure for the Lifetime Performance Index119862119871with the Bayesian Estimator under General Entropy Loss
Function Let 119883 be the lifetime of such a product and 119883has a Rayleigh distribution with the pdf as given by (2)We consider the general entropy (GE) loss function (also see[12 20])
119871 (120579lowast 120579) prop (
120579lowast
120579)
119902
minus 119902 log(120579lowast
120579) minus 1 (37)
whose minimum occurs at 120579lowast = 120579The loss function is a generalization of entropy loss used
by several authors (eg Dyer and Liu [24] Soliman [25] andSoliman and Elkahlout [26]) where the shape parameter 119902 =1 When 119902 gt 0 a positive error (120579lowast gt 120579) causes more seriousconsequences than a negative error The Bayesian estimator120579BG of 120579 under GE loss function is given by
120579BG = [119864 (120579minus119902| 119909119880(1) 119909
119880(119899))]minus1119902
(38)
By using (18) and (38) then the Bayesian estimator 120579BG of120579 is derived as follows
119864 (120579minus119902| 119909119880(1) 119909
119880(119899))
= int
infin
0
120579minus1199021199011(120579 | 119909
119880(1) 119909
119880(119899)) 119889120579
= (119905
2)
minus1199022
sdotΓ (119899 + 119887 + 1199022)
Γ (119899 + 119887)
(39)
where 119905 = 1199092119880(119899)
+ 119886 Hence the Bayesian estimator 120579BG of 120579under the GE loss function is given by
120579BG = (119879
2)
12
[Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(40)
where 119879 = 1198832
119880(119899)+ 119886 119886 and 119887 are the parameters of prior
distribution with density as (15)By using (5) and the Bayesian estimator 120579BG as (40)119862
119871BGbased on the Bayesian estimator 120579BG of 120579 is given by
119862119871BG = radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579BG
= radic120587
4 minus 120587minus radic
2
4 minus 120587sdot 119871
sdot [(119879
2)
minus1199022Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
1119902
(41)
8 Journal of Applied Mathematics
where 119879 = 1198832
119880(119899)+ 119886 119886 and 119887 are the parameters of prior
distribution with density as (15)We construct a statistical testing procedure to assess
whether the lifetime performance index adheres to therequired level The one-sided credible confidence interval for119862119871is obtained using the pivotal quantity 1199051205792|
119909sim
=(119909119880(1)119909119880(119899))
By using the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899)) given the
specified significance level 120572 the level 100(1minus120572) one-sidedcredible interval for 119862
119871can be derived as follows
Since the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899))sim 1205942
2(119899+119887)
and 12059422(119899+119887)1minus120572
function which represents the lower 1 minus 120572percentile of 1205942
2(119899+119887) then
119875(119905
1205792le 1205942
2(119899+119887)1minus120572| 119909sim) = 1 minus 120572
where 119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899))
997904rArr 119875(119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times[Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
| 119909sim)
= 1 minus 120572
(42)
where 119862119871as the definition of (5)
From (42) we obtain that a 100(1 minus 120572) one-sidedcredible interval for 119862
119871is
119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times [Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
(43)
where 119862119871BG is given by (41)
Therefore the 100(1 minus 120572) lower credible interval boundfor 119862119871can be written as
LBBG = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times [Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
(44)
where 119862119871BG is given by (41) 120572 is the specified significance
level and 119887 is a parameter of prior distribution with densityas (15)
The managers can use the one-sided credible interval todetermine whether the product performance attains to therequired level The proposed testing procedure of 119862
119871with
119862119871BG can be organized as follows
Step 1 Determine the lower lifetime limit 119871 for productsand the performance index value 119888
0 then the testing null
hypothesis 1198670 119862119871le 1198880and the alternative hypothesis
1198671 119862119871gt 1198880are constructed
Step 2 Specify a significance level 120572
Step 3 Given the parameters 119886 and 119887 of prior distributionthe parameter 119902 of GE loss function the upper record values119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899)) the lower lifetime limit 119871 and the
significance level 120572 then we can calculate the 100(1 minus 120572)one-sided credible interval [LBBGinfin) for 119862119871 where LBBG asthe definition of (44)
Step 4 The decision rule of statistical test is provided asfollows
(1) if the performance index value 1198880notin [LBBGinfin) then
we will reject 1198670 It is concluded that the lifetime
performance index of products meets the requiredlevel
(2) if the performance index value 1198880isin [LBBGinfin) then
we do not reject 1198670 It is concluded that the lifetime
performance index of products does not meet therequired level
6 The Monte Carlo Simulation Algorithm ofConfidence (or Credible) Level
In this section we will report the results of a simulationstudy for confidence (or credible) level (1 minus 120572) based on a100(1 minus 120572) one-sided confidence (or credible) interval ofthe lifetime performance index 119862
119871 We considered 120572 = 005
and then generated samples from the Rayleigh distributionwith pdf as in (2) with respect to the record values
(i) The Monte Carlo simulation algorithm of confidencelevel (1 minus 120572) for 119862
119871under MLE is given in the
following steps
Step 1 Given 119899 119886 119887 119871 and 120572 where 119886 gt 0 119887 gt 0
Step 2 For the values of prior parameters (119886 119887) use (15) togenerate 120579 from the square-root inverted-gamma distribu-tion
Step 3(a) The generation of data (119885
1 1198852 119885
119899) is by standard
exponential distribution with parameter 1(b) Set1198842
1= 212057921198851and1198842119894= 21205792119885119894+1198842
119894minus1 for 119894 = 2 119899
1198841= radic21205792119885
1 119884
119894= radic21205792119885
119894+ 1198842
119894minus1 for 119894 = 2 119899
(45)
(1198841 1198842 119884
119899) are the upper record values from the
Rayleigh distribution(c) The value of LBMLE is calculated by
LBMLE = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871MLE)(
1205942
21198991minus120572
2119899)
12
(46)
Journal of Applied Mathematics 9
where 119862119871MLE is given by (11) and 1205942
21198991minus120572function
which represents the lower 1 minus 120572 percentile of 12059422119899
(d) If 119862119871ge LBMLE then count = 1 else count = 0
Step 4
(a) Step 3 is repeated 1000 times(b) The estimation of confidence level (1 minus 120572) is (1 minus 120572) =
(total count)1000
Step 5
(a) Repeat Step 2ndashStep 4 for 100 times then we can getthe 100 estimations of confidence level as follows
(1 minus 120572)1 (1 minus 120572)
2 (1 minus 120572)
100 (47)
(b) The average empirical confidence level 1 minus 120572
of (1 minus 120572)119894 119894 = 1 100 that is 1 minus 120572 =
(1100)sum100
119894=1(1 minus 120572)
119894
(c) The sample mean square error (SMSE) of(1 minus 120572)
1 (1 minus 120572)
2 (1 minus 120572)
100SMSE = (1100)
times sum100
119894=1[(1 minus 120572)
119894minus (1 minus 120572)]
2
The results of simulation are summarized in Table 2 basedon 119871 = 10 the different value of sample size 119899 priorparameter (119886 119887) = (2 5) (6 15) and (2 2) at 120572 = 005respectively The scope of SMSE is between 000413 and000593
Step 1 Given 119899 119886 119887 119871 and 120572 where 119886 gt 0 119887 gt 0
Step 2 For the values of prior parameters (119886 119887) use (15) togenerate 120579 from the square-root inverted-gamma distribu-tion
Step 3
(a) The generation of data (1198851 1198852 119885
119899) is by standard
exponential distribution with parameter 1(b) Set1198842
1= 212057921198851and1198842119894= 21205792119885119894+1198842
119894minus1 for 119894 = 2 119899
1198841= radic21205792119885
1 119884
119894= radic21205792119885
119894+ 1198842
119894minus1 for 119894 = 2 119899
(48)
(1198841 1198842 119884
119899) are the upper record values from the
Rayleigh distribution(c) The values of LBBS LBBL and LBBG are calculated by
(25) (36) and (44) respectively(d) If119862
119871ge LB then count = 1 else count = 0 where LB =
LBBS or LBBL or LBBG
Step 4
(a) Step 3 is repeated 1000 times(b) The estimation of credible level (1 minus 120572) is (1 minus 120572) =
(total count)1000
Step 5
(a) Repeat Step 2ndashStep 4 for 100 times then we can getthe 100 estimations of credible level as follows
(1 minus 120572)1 (1 minus 120572)
2 (1 minus 120572)
100 (49)
(b) The average empirical credible level 1 minus 120572 of (1 minus 120572)119894
119894 = 1 100 that is 1 minus 120572 = (1100)sum100119894=1(1 minus 120572)
119894
(c) The sample mean square error (SMSE) of(1 minus 120572)
1 (1 minus 120572)
2 (1 minus 120572)
100SMSE = (1100)
times sum100
119894=1[(1 minus 120572)
119894minus (1 minus 120572)]
2
Based on 119871 = 10 the different value of sample size119899 prior parameter (119886 119887) = (2 5) (6 15) (2 2) and 120572 =
005 the results of simulation are summarized in Tables 3ndash5under SE loss function LINEX loss function with parameter119888 = minus05 05 15 and GE loss function with parameter119902 = 3 5 8 respectively The following points can be drawn
(a) All of the SMSEs are small enough and the scope ofSMSE is between 000346 and 000467
(b) Fix the size 119899 comparison of prior parameter (119886 119887) =(2 5) (6 15) and (2 2) as follows
(i) the values of SMSEwith prior parameter (119886 119887) =(6 15) and (2 5) are smaller than (119886 119887) = (2 2)respectively
(ii) a comparison of prior parameter (119886 119887) = (2 5)and (2 2) that is fix prior parameter 119886 if priorparameter 119887 increases then it can be seen thatthe SMSE will decrease
(c) In Table 4 fix the size 119899 and the prior parameter (119886 119887)comparison the parameter of LINEX loss function 119888 =minus05 05 15 as followsthe values of SMSE with prior parameter 119888 = 05 15andminus15 are the same as the values of SMSE inTable 3
(d) In Table 5 fix the prior parameter (119886 119887) comparisonof GE loss function parameter 119902 = 3 5 8 as followsthe values of SMSE with prior parameter 119902 = 3 5 and8 are the same as the values of SMSE in Table 3
Hence these results from simulation studies illustrate thatthe performance of our proposed method is acceptableMoreover we suggest that the prior parameter (119886 119887) = (6 15)and (2 5) is appropriate for the square-root inverted-gammadistribution
7 Numerical Example
In this section we propose the new hypothesis testingprocedures to a real-life data In the following examplewe discuss a real-life data for 25 ball bearings to illustratethe use of the new hypothesis testing procedures in thelifetime performance of ball bearings The proposed testingprocedures not only can handle nonnormal lifetime data butalso can handle the upper record values
10 Journal of Applied Mathematics
Table 2 Average empirical confidence level (1 minus 120572) for 119862119871under MLE when 120572 = 005
119899 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
5 093653 (000462) 093482 (000467) 093577 (000507)10 093643 (000449) 093581 (000413) 093450 (000593)15 093523 (000489) 093557 (000503) 093463 (000516)119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 3 Average empirical credible level (1 minus 120572) for 119862119871under SE loss function when 120572 = 005
119899 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
5 094357 (000386) 093760 (000386) 093925 (000395)10 094244 (000359) 093704 (000359) 093664 (000462)15 094096 (000346) 093619 (000426) 093635 (000467)119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 4 Average empirical credible level (1 minus 120572) for 119862119871under LINEX loss function when 120572 = 005
119899 119888 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
minus05 094357 (000386) 093760 (000386) 093925 (000395)5 05 094357 (000386) 093760 (000386) 093925 (000395)
15 094357 (000386) 093760 (000386) 093925 (000395)minus05 094244 (000359) 093704 (000359) 093664 (000462)
10 05 094244 (000359) 093704 (000359) 093664 (000462)15 094244 (000359) 093704 (000359) 093664 (000462)minus05 094096 (000346) 093619 (000426) 093635 (000467)
15 05 094096 (000346) 093619 (000426) 093635 (000467)15 094096 (000346) 093619 (000426) 093635 (000467)
119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 5 Average empirical credible level (1 minus 120572) for 119862119871under GE loss function when 120572 = 005
119899 119902 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
3 094357 (000386) 093760 (000386) 093925 (000395)5 5 094357 (000386) 093760 (000386) 093925 (000395)
8 094357 (000386) 093760 (000386) 093925 (000395)3 094244 (000359) 093704 (000359) 093664 (000462)
10 5 094244 (000359) 093704 (000359) 093664 (000462)8 094244 (000359) 093704 (000359) 093664 (000462)3 094096 (000346) 093619 (000426) 093635 (000467)
15 5 094096 (000346) 093619 (000426) 093635 (000467)8 094096 (000346) 093619 (000426) 093635 (000467)
119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Example The data is the failure times of 25 ball bearings inendurance testThe 25 observations are the number ofmillionrevolutions before failure for each of the ball bearings Thedata come from Caroni [27] as follows
6780 6780 6780 6864 3300 6864 9864 12804 42122892 4560 5184 5556 17340 4848 1788 9312 54124152 5196 12792 8412 10512 10584 and 6888
Raqab and Madi [28] and Lee [29] indicated that a one-parameter Rayleigh distribution is acceptable for these dataIn addition we also have a way to test the hypothesis thatthe failure data come from the Rayleigh distribution Thetesting hypothesis119867
0 119883 comes from a Rayleigh distribution
versus1198671 119883 does not come from a Rayleigh distribution is
constructed under significance level 120572 = 005From a probability plot of Figure 1 by using the Minitab
Statistical Software we can conclude the operational lifetimesdata of 25 ball bearings from the Rayleigh distribution whichis the Weibull distribution with the shape 2 (also see Lee etal [14]) For informative prior we use the prior information119864(120579) = 120579MLE ≐ 54834 and Var(120579) = 02 giving the priorparameter values as (119886 = 6014 119887 = 1001)
Here if only the upper record values have been observedthese are 119909
119880(119894) 119894 = 1 5(= 119899) = 6780 6864 9864
12804 17340
Journal of Applied Mathematics 11
10010
99
90807060504030
20
10
5
32
1
Xi
Probability plot of XiWeibull-95 CI
()
Shape 2
Scale 8002N 25
AD 0431P -value 057
Figure 1 Probability plot for failure times of 25 ball bearings data
(1) The proposed testing procedure of 119862119871with 119862
119871MLE isstated as follows
Step 1 The record values from the above data are (119909119880(119894) 119894 =
1 2 3 4 5) = (6780 6864 9864 12804 17340)
Step 2 The lower specification limit 119871 is assumed to be 2337The deal with the product managersrsquo concerns regardinglifetime performance and the conforming rate 119875
119903of products
is required to exceed 80 percent Referring to Table 1 119862119871is
required to exceed 090Thus the performance index value isset at 119888
0= 090 The testing hypothesis119867
0 119862119871le 090 versus
1198671 119862119871gt 090 is constructed
Step 3 Specify a significance level 120572 = 005
Step 4 We can calculate that the 95 lower confidenceinterval bound for 119862
119871 where
LBMLE = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871MLE)(
1205942
21198991minus120572
2119899)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 126251254)(
1205942
10095
10)
12
= 103285
(50)So the 95 one-sided confidence interval for 119862
119871is
[LBMLEinfin) = [103285infin)
Step 5 Because of the performance index value 1198880= 090 notin
[LBMLEinfin) we reject the null hypothesis1198670 119862119871 le 090
Thus we can conclude that the lifetime performanceindex of data meets the required level
(2) The proposed testing procedures of 119862119871with 119862
119871BS119862119871BL and 119862119871BG are stated as follows
Step 1 The record values from the above data are (119909119880(119894) 119894 =
1 2 3 4 5) = (6780 6864 9864 12804 17340)
Step 2 The lower specification limit 119871 is assumed to be 2337The deal with the product managersrsquo concerns regardinglifetime performance and the conforming rate 119875
119903of products
is required to exceed 80 percent Referring to Table 1 119862119871is
required to exceed 090Thus the performance index value isset at 119888
0= 090 The testing hypothesis119867
0 119862119871le 090 versus
1198671 119862119871gt 090 is constructed
Step 3 Specify a significance level 120572 = 005 the parameter ofLINEX loss function 119888 = 05 and the parameter of GE lossfunction 119902 = 2
Step 4 We can calculate the 95 lower credible intervalbound for 119862
119871 where
LBBS = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BS)
timesΓ (119899 + 119887 minus 12)
Γ (119899 + 119887)(1205942
2(119899+119887)1minus120572
2)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1246074685)
timesΓ (5 + 1001 minus 12)
Γ (5 + 1001)(1205942
2(5+1001)095
2)
12
= 096984
(51)
12 Journal of Applied Mathematics
So the 95 one-sided credible interval for 119862119871
is[LBBSinfin) = [096984infin)
LBBL = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1129561059)
times [1
2 (05)(1 minus 119890
minus05(5+1001+1)) 1205942
2(5+1001)095]
12
= 096984
(52)
So the 95 one-sided credible interval for 119862119871
is[LBBLinfin) = [096984infin)
LBBG = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times [Γ (119899 + 119887 + 1199022)
Τ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1200432998)
times [Γ (5 + 1001 + 22)
Γ (5 + 1001)]
minus12
(1205942
2(5+1001)095
2)
12
= 096984
(53)
So the 95 one-sided credible interval for 119862119871
is[LBBGinfin) = [096984infin)
Step 5 Because of the performance index value 1198880= 090 notin
[LBBSinfin) = [LBBLinfin) = [LBBGinfin) we reject the nullhypothesis119867
0 119862119871le 090
Thus we can conclude that the lifetime performance index ofdata meets the required level In addition by using the priorparameter values (119886 119887) = (6 15) based on the simulationstudies we also obtain that 119888
0= 090 notin [LBBSinfin) =
[LBBLinfin) = [LBBGinfin) = 094033 So we also rejectthe null hypothesis 119867
0 119862119871le 090 Hence the lifetime
performance index of data meets the required level
8 Conclusions
Montgomery [1] proposed a process capability index 119862119871for
larger-the-better quality characteristic The assumption ofmost process capability is normal distribution but it is ofteninvalid In this paper we consider that Rayleigh distributionis the special case of Weibull distribution
Moreover in lifetime data testing experiments the exper-imenter may not always be in a position to observe the life
times of all the products put to test In industry and reliabilitystudies many products fail under stress for example anelectronic component ceases to function in an environmentof too high temperature and a battery dies under the stressof time So in such experiments measurement may be madesequentially and only the record values are observed
In order to let the process capability indices can beeffectively used This study constructs MLE and Bayesianestimator of 119862
119871under assuming the conjugate prior distribu-
tion and SE loss function LINEX loss function and GE lossfunction based on the upper record values from the Rayleighdistribution The Bayesian estimator of 119862
119871is then utilized to
develop a credible interval in the condition of known 119871This study also provides a table of the lifetime perfor-
mance index with its corresponding conforming rate basedon the Rayleigh distributionThat is the conforming rate andthe corresponding 119862
119871can be obtained
Further the Bayesian estimator and posterior distributionof 119862119871are also utilized to construct the testing procedure of
119862119871which is based on a credible interval If you want to test
whether the products meet the required level you can utilizethe proposed testing procedure which is easily applied andan effectively method to test in the condition of known 119871Numerical example is illustrated to show that the proposedtesting procedure is effectively evaluating whether the trueperformance index meets requirements
In addition these results from simulation studies illus-trate that the performance of our proposedmethod is accept-able According to SMSE the Bayesian approach is smallerthan the non-Bayesian approach we suggest that the Bayesianapproach is better than the non-Bayesian approachThen theSMSEs of prior parameters (119886 119887) = (6 15) and (2 5) aresmaller than the SMSE of prior parameter (119886 119887) = (2 2)We suggest that the prior parameters (119886 119887) = (6 15) and(2 5) are appropriate for the square-root inverted-gammadistribution based on the Bayesian estimators under theRayleigh distribution
Acknowledgments
The authors thank the referees for their suggestions andhelpful comments on revising the paper This research waspartially supported by the National Science Council China(Plans nos NSC 101-2221-E-158-003 NSC 99-2118-M-415-001 and NSC 100-2118-M-415-002) Moreover J W Wu andC W Hong have a direct financial relation with the NationalScience Council Taiwan but other authors do not a directfinancial relation with the National Science Council Taiwan
References
[1] D C Montgomery Introduction To Statistical Quality ControlJohn Wiley amp Sons New York NY USA 1985
[2] L I Tong K S Chen and H T Chen ldquoStatistical testingfor assessing the performance of lifetime index of electroniccomponents with exponential distributionrdquo Journal of QualityReliability Management vol 19 pp 812ndash824 2002
[3] C-W Hong J-W Wu and C-H Cheng ldquoComputationalprocedure of performance assessment of lifetime index of
Journal of Applied Mathematics 13
businesses for the Pareto lifetime model with the right type IIcensored samplerdquo Applied Mathematics and Computation vol184 no 2 pp 336ndash350 2007
[4] C W Hong J W Wu and C H Cheng ldquoComputational pro-cedure of performance assessment of lifetime index of Paretolifetime businesses based on confidence intervalrdquo Applied SoftComputing vol 8 no 1 pp 698ndash705 2008
[5] J-W Wu H-M Lee and C-L Lei ldquoComputational testingalgorithmic procedure of assessment for lifetime performanceindex of products with two-parameter exponential distribu-tionrdquo Applied Mathematics and Computation vol 190 no 1 pp116ndash125 2007
[6] W-C Lee J-W Wu and C-W Hong ldquoAssessing the lifetimeperformance index of products with the exponential distribu-tion under progressively type II right censored samplesrdquo Journalof Computational and Applied Mathematics vol 231 no 2 pp648ndash656 2009
[7] W C Lee J W Wu and C L Lei ldquoEvaluating the lifetimeperformance index for the exponential lifetime productsrdquoApplied Mathematical Modelling vol 34 no 5 pp 1217ndash12242010
[8] K N Chandler ldquoThe distribution and frequency of recordvaluesrdquo Journal of the Royal Statistical Society B vol 14 pp 220ndash228 1952
[9] S K Bhattacharya and R K Tyagi ldquoBayesian survival analysisbased on the rayleigh modelrdquo Trabajos de Estadistica vol 5 no1 pp 81ndash92 1990
[10] A D Cliff and J K Ord ldquoModel building and the analysis ofspatial pattern in human geography (with discussion)rdquo Journalof the Royal Statistical Society B vol 37 no 3 pp 297ndash348 1975
[11] D D Dyer and C W Whisenand ldquoBest linear unbiasedestimator of the parameter of the Rayleigh distribution I Smallsample theory for censored order statisticsrdquo IEEE Transactionson Reliability vol R-22 no 1 pp 27ndash34 1973
[12] A A Soliman and F M Al-Aboud ldquoBayesian inference usingrecord values from Rayleigh model with applicationrdquo EuropeanJournal of Operational Research vol 185 no 2 pp 659ndash6722008
[13] StatSoft STATISTICA Software Version 7 StatSoft Tulsa Okla2005
[14] W-C Lee J-W Wu M-L Hong L-S Lin and R-L ChanldquoAssessing the lifetime performance index of Rayleigh productsbased on the Bayesian estimation under progressive type IIright censored samplesrdquo Journal of Computational and AppliedMathematics vol 235 no 6 pp 1676ndash1688 2011
[15] M Ahsanullah Record Statistics Nova Science New York NYUSA 1995
[16] B C Arnold N Balakrishnan and H N Nagaraja RecordsJohn Wiley amp Sons New York NY USA 1998
[17] P W Zehna ldquoInvariance of maximum likelihood estimatorsrdquoAnnals of Mathematical Statistics vol 37 p 744 1966
[18] A Zellner ldquoBayesian estimation and prediction using asymmet-ric loss functionsrdquo Journal of the American Statistical Associa-tion vol 81 no 394 pp 446ndash451 1986
[19] S-Y Huang ldquoEmpirical Bayes testing procedures in somenonexponential families using asymmetric linex loss functionrdquoJournal of Statistical Planning and Inference vol 46 no 3 pp293ndash309 1995
[20] R Calabria andG Pulcini ldquoPoint estimation under asymmetricloss functions for left-truncated exponential samplesrdquo Commu-nications in StatisticsmdashTheory and Methods vol 25 no 3 pp585ndash600 1996
[21] A T K Wan G Zou and A H Lee ldquoMinimax and Γ-minimaxestimation for the Poisson distribution under LINEX loss whenthe parameter space is restrictedrdquo Statistics and ProbabilityLetters vol 50 no 1 pp 23ndash32 2000
[22] S Anatolyev ldquoKernel estimation under linear-exponential lossrdquoEconomics Letters vol 91 no 1 pp 39ndash43 2006
[23] X Li Y Shi J Wei and J Chai ldquoEmpirical Bayes estimators ofreliability performances using LINEX loss under progressivelytype-II censored samplesrdquo Mathematics and Computers inSimulation vol 73 no 5 pp 320ndash326 2007
[24] D K Dyer and P S L Liu ldquoOn comparison of estimators in ageneralized life modelrdquo Microelectronics Reliability vol 32 no1-2 pp 207ndash221 1992
[25] A A Soliman ldquoEstimation of parameters of life from progres-sively censored data using Burr-XII modelrdquo IEEE Transactionson Reliability vol 54 no 1 pp 34ndash42 2005
[26] A A Soliman and G R Elkahlout ldquoBayes estimation of thelogistic distribution based on progressively censored samplesrdquoJournal of Applied Statistical Science vol 14 no 3-4 pp 281ndash2932005
[27] C Caroni ldquoThe correct ldquoball bearingsrdquo datardquo Lifetime DataAnalysis vol 8 no 4 pp 395ndash399 2002
[28] M Z Raqab and M T Madi ldquoBayesian prediction of the totaltime on test using doubly censored Rayleigh datardquo Journal ofStatistical Computation and Simulation vol 72 no 10 pp 781ndash789 2002
[29] W C Lee ldquoStatistical testing for assessing lifetime performanceIndex of the Rayleigh lifetime productsrdquo Journal of the ChineseInstitute of Industrial Engineers vol 25 pp 433ndash445 2008
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Journal of Applied Mathematics 5
as follows
1205871 (120579) =
119886119887
Γ (119887) 2119887minus1120579minus2119887minus1
119890minus1198862120579
2
(15)
where 120579 gt 0 119886 gt 0 and 119887 gt 0
119864 (120579) = (119886
2)
12Γ (119887 minus 12)
Γ (119887)
Var (120579) = 119886
2 (119887 minus 1)minus (
119886
2) [Γ (119887 minus 12)
Γ (119887)]
2
119887 gt 1
(16)
With record values 119899 products (or items) take place ontest Consider (119883
119880(1) 119883119880(2) 119883
119880(119899)) is the upper record
values We can obtain that the posterior pdf of 120579 | (119883119880(1)
119883119880(2) 119883
119880(119899)) is given as
1199011(120579 | 119909
119880(1) 119909
119880(119899))
=119871 (120579 | 119909
119880(1) 119909
119880(119899)) 1205871 (120579)
intinfin
0119871 (120579 | 119909
119880(1) 119909
119880(119899)) 1205871 (120579) 119889120579
=
120579minus2(119899+119887)minus1
(1199092
119880(119899)+ 119886)(119899+119887)
119890minus(119909119880(119899)+119886)2120579
2
Γ (119899 + 119887) 2119899+119887minus1
(17)
where 120579 gt 0 and the likelihood function 119871(120579 | 119909119880(1)
119909119880(119899)) as (8)Let 119905 = 1199092
119880(119899)+119886 then the posterior pdf can be rewritten
as
1199011(120579 | 119909
119880(1) 119909
119880(119899)) =
120579minus2(119899+119887)minus1
119905(119899+119887)
119890minus1199052120579
2
Γ (119899 + 119887) 2119899+119887minus1 (18)
and we can show that 1199051205792|119909sim
=(119909119880(1)119909119880(119899))sim 1205942
2(119899+119887)
The Bayesian estimator 120579BS of 120579 based on the squarederror (SE) loss function 119871(120579lowast 120579) = (120579lowast minus120579)2 can be derived asfollows
119864 (120579 | 119909119880(1) 119909
119880(119899))
= int
infin
0
1205791199011(120579 | 119909
119880(1) 119909
119880(119899)) 119889120579
=Γ (119899 + 119887 minus 12)
Γ (119899 + 119887)(119905
2)
12
(19)
where 119905 = 1199092119880(119899)
+ 119886 119899 + 119887 minus 12 gt 0Hence the Bayesian estimator of 120579 based on SE loss
function is given by
120579BS =Γ (119899 + 119887 minus 12)
Γ (119899 + 119887)(119879
2)
12
(20)
where 119879 = 1198832
119880(119899)+ 119886 119899 + 119887 minus 12 gt 0 119886 and 119887 are the
parameters of prior distribution with density as (15)The lifetime performance index 119862
119871of Rayleigh products
can be written as
119862119871= radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579 minusinfin lt 119862
119871lt radic
120587
4 minus 120587 (21)
By using (5) and the Bayesian estimator 120579BS as (20) 119862119871BS
based on the Bayesian estimator 120579BS of 120579 is given by
119862119871BS = radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579BS
= radic120587
4 minus 120587minus radic
2
4 minus 120587sdot 119871 sdot [
Γ (119899 + 119887 minus 12)
Γ (119899 + 119887)(119879
2)
12
]
minus1
(22)
where 119879 = 1198832
119880(119899)+ 119886 119899 + 119887 minus 12 gt 0 119886 and 119887 are the
parameters of prior distribution with density as in (15)We construct a statistical testing procedure to assess
whether the lifetime performance index adheres to therequired level The one-sided credible confidence interval for119862119871are obtained using the pivotal quantity 1199051205792|
119909sim
=(119909119880(1)119909119880(119899))
By using the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899)) given the
specified significance level 120572 the level 100(1minus120572) one-sidedcredible interval for 119862
119871can be derived as follows
Since the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899))sim 1205942
2(119899+119887)
and 12059422(119899+119887)1minus120572
function which represents the lower 1 minus 120572percentile of 1205942
2(119899+119887) then
119875(119905
1205792le 1205942
2(119899+119887)1minus120572| 119909sim) = 1 minus 120572
where 119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899))
997904rArr 119875(119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BS)
timesΓ (119899 + 119887 minus 12)
Γ (119899 + 119887)(1205942
2(119899+119887)1minus120572
2)
12
| 119909sim)
= 1 minus 120572
(23)
where 119862119871as the definition of (5)
From (23) we obtain that a 100(1minus120572)one-sided credibleinterval for 119862
119871is given by
119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BS)
timesΓ (119899 + 119887 minus 12)
Γ (119899 + 119887)(1205942
2(119899+119887)1minus120572
2)
12
(24)
where 119862119871BS is given by (22)
Therefore the 100(1 minus 120572) lower credible interval boundfor 119862119871can be written as
LBBS = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BS)
timesΓ (119899 + 119887 minus 12)
Γ (119899 + 119887)(1205942
2(119899+119887)1minus120572
2)
12
(25)
6 Journal of Applied Mathematics
where 119862119871BS is given by (22) and 119887 is a parameter of prior
distribution with density as (15)The managers can use the one-sided credible interval to
determine whether the product performance adheres to therequired level The proposed testing procedure of 119862
119871with
119862119871BS can be organized as follows
Step 1 Determine the lower lifetime limit 119871 for productsand the performance index value 119888
0 then the testing null
hypothesis 1198670 119862119871le 1198880and the alternative hypothesis
1198671 119862119871gt 1198880are constructed
Step 2 Specify a significance level 120572
Step 3 Given the parameters 119886 and 119887 of prior distributionthe upper record values 119909
sim= (119909
119880(1) 119909119880(2) 119909
119880(119899)) the
lower lifetime limit 119871 and the significance level 120572 then wecan calculate the 100(1 minus 120572) one-sided credible interval[LBBSinfin) for 119862119871 where LBBS as the definition of (25)
Step 4 The decision rule of statistical test is provided asfollows
(1) if the performance index value 1198880notin [LBBSinfin) then
we will reject 1198670 It is concluded that the lifetime
performance index of products meets the requiredlevel
(2) if the performance index value 1198880isin [LBBSinfin) then
we do not reject 1198670 It is concluded that the lifetime
performance index of products does not meet therequired level
52 Testing Procedure for the Lifetime Performance Index 119862119871
with the Bayesian Estimator under Linear Exponential LossFunction Let 119883 be the lifetime of such a product and 119883has a Rayleigh distribution with the pdf as given by (2) Weconsider the linear exponential (LINEX) loss function as (alsosee [18ndash23])
119871 (Δ) prop 119890119888Δminus 119888Δ minus 1 (26)
where Δ = (120579lowast minus 120579) 119888 = 0In this paper we suppose that
Δ1= (
120579lowast
120579)
2
minus 1 (27)
where 120579lowast is an estimator of 120579By using (26)-(27) the posterior expectation of LINEX
loss function 119871(Δ1) is
119864 [119871 (Δ1) | 119909sim] = 119890minus119888119864 [119890119888(120579lowast
120579)2 | 119909sim]
minus 119888119864 [(120579lowast
120579)
2
minus 1 | 119909sim] minus 1
(28)
where 119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899))
The value of 120579lowast that minimizes 119864[119871(Δ1) | 119909sim] denoted by
120579BL is obtained by solving the equation
119889119864 [119871 (Δ1) | 119909sim]
119889120579lowast= 119890minus119888119864[119890119888(120579lowast
120579)2 sdot2119888120579lowast
1205792| 119909sim]
minus 119888119864(2120579lowast
1205792| 119909sim) = 0
(29)
that is 120579BL is the solution to the following equation
119864[120579BL1205792119890119888(120579BL120579)2 | 119909
sim] = 119890119888119864(
120579BL1205792
| 119909sim) (30)
By using (18) and (30) we have
2120579BL (119899 + 119887) sdot119905119899+119887
(119905 minus 21198881205792
BL)119899+119887+1
= 119890119888120579BL
2 (119899 + 119887)
119905
where 119905 = 1199092119880(119899)
+ 119886
(31)
Hence the Bayesian estimator of 120579 under the LINEX lossfunction is given by
120579BL = [119879
2119888(1 minus 119890
minus119888(119899+119887+1))]
12
(32)
where 119879 = 1198832
119880(119899)+ 119886 119886 and 119887 are the parameters of prior
distribution with density as (15)By using (5) and the Bayesian estimator 120579BL as (32) 119862119871BL
based on the Bayesian estimator 120579BL of 120579 is given by
119862119871BL = radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579BL
= radic120587
4 minus 120587minus radic
2
4 minus 120587119871[119879
2119888(1 minus 119890
minus119888(119899+119887+1))]
minus12
(33)
where 119879 = 1198832
119880(119899)+ 119886 119886 and 119887 are the parameters of prior
distribution with density as (15)We construct a statistical testing procedure to assess
whether the lifetime performance index adheres to therequired level The one-sided credible confidence interval for119862119871is obtained using the pivotal quantity 1199051205792|
119909sim
=(119909119880(1)119909119880(119899))
By using the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899)) given the
specified significance level 120572 the level 100(1minus120572) one-sidedcredible interval for 119862
119871can be derived as follows
Journal of Applied Mathematics 7
Since the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899))sim 1205942
2(119899+119887)
and 12059422(119899+119887)1minus120572
function which represents the lower 1 minus 120572percentile of 1205942
2(119899+119887) then
119875(119905
1205792le 1205942
2(119899+119887)1minus120572| 119909sim) = 1 minus 120572
where 119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899))
997904rArr 119875(119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
| 119909sim)
= 1 minus 120572
(34)
where 119862119871as the definition of (5)
From (34) we obtain that a 100(1minus120572)one-sided credibleinterval for 119862
119871is
119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
(35)
where 119862119871BL is given by (33)
Therefore the 100(1 minus 120572) lower credible interval boundfor 119862119871can be written as
LBBL = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
(36)
where 119862119871BL is given by (33) and 119887 is a parameter of prior
distribution with density as (15)The managers can use the one-sided credible interval to
determine whether the product performance attains to therequired level The proposed testing procedure of 119862
119871with
119862119871BL can be organized as follows
Step 1 Determine the lower lifetime limit 119871 for productsand the performance index value 119888
0 then the testing null
hypothesis 1198670 119862119871le 1198880and the alternative hypothesis
1198671 119862119871gt 1198880are constructed
Step 2 Specify a significance level 120572
Step 3 Given the parameters 119886 and 119887 of prior distribution theparameter 119888 of LINEX loss function the upper record values119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899)) the lower lifetime limit 119871 and the
significance level 120572 then we can calculate the 100(1 minus 120572)one-sided credible interval [LBBLinfin) for 119862119871 where LBBL asthe definition of (36)
Step 4 The decision rule of statistical test is provided asfollows
(1) if the performance index value 1198880notin [LBBLinfin) then
we will reject 1198670 It is concluded that the lifetime
performance index of products meets the requiredlevel
(2) if the performance index value 1198880isin [LBBLinfin) then
we do not reject 1198670 It is concluded that the lifetime
performance index of products does not meet therequired level
53 Testing Procedure for the Lifetime Performance Index119862119871with the Bayesian Estimator under General Entropy Loss
Function Let 119883 be the lifetime of such a product and 119883has a Rayleigh distribution with the pdf as given by (2)We consider the general entropy (GE) loss function (also see[12 20])
119871 (120579lowast 120579) prop (
120579lowast
120579)
119902
minus 119902 log(120579lowast
120579) minus 1 (37)
whose minimum occurs at 120579lowast = 120579The loss function is a generalization of entropy loss used
by several authors (eg Dyer and Liu [24] Soliman [25] andSoliman and Elkahlout [26]) where the shape parameter 119902 =1 When 119902 gt 0 a positive error (120579lowast gt 120579) causes more seriousconsequences than a negative error The Bayesian estimator120579BG of 120579 under GE loss function is given by
120579BG = [119864 (120579minus119902| 119909119880(1) 119909
119880(119899))]minus1119902
(38)
By using (18) and (38) then the Bayesian estimator 120579BG of120579 is derived as follows
119864 (120579minus119902| 119909119880(1) 119909
119880(119899))
= int
infin
0
120579minus1199021199011(120579 | 119909
119880(1) 119909
119880(119899)) 119889120579
= (119905
2)
minus1199022
sdotΓ (119899 + 119887 + 1199022)
Γ (119899 + 119887)
(39)
where 119905 = 1199092119880(119899)
+ 119886 Hence the Bayesian estimator 120579BG of 120579under the GE loss function is given by
120579BG = (119879
2)
12
[Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(40)
where 119879 = 1198832
119880(119899)+ 119886 119886 and 119887 are the parameters of prior
distribution with density as (15)By using (5) and the Bayesian estimator 120579BG as (40)119862
119871BGbased on the Bayesian estimator 120579BG of 120579 is given by
119862119871BG = radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579BG
= radic120587
4 minus 120587minus radic
2
4 minus 120587sdot 119871
sdot [(119879
2)
minus1199022Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
1119902
(41)
8 Journal of Applied Mathematics
where 119879 = 1198832
119880(119899)+ 119886 119886 and 119887 are the parameters of prior
distribution with density as (15)We construct a statistical testing procedure to assess
whether the lifetime performance index adheres to therequired level The one-sided credible confidence interval for119862119871is obtained using the pivotal quantity 1199051205792|
119909sim
=(119909119880(1)119909119880(119899))
By using the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899)) given the
specified significance level 120572 the level 100(1minus120572) one-sidedcredible interval for 119862
119871can be derived as follows
Since the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899))sim 1205942
2(119899+119887)
and 12059422(119899+119887)1minus120572
function which represents the lower 1 minus 120572percentile of 1205942
2(119899+119887) then
119875(119905
1205792le 1205942
2(119899+119887)1minus120572| 119909sim) = 1 minus 120572
where 119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899))
997904rArr 119875(119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times[Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
| 119909sim)
= 1 minus 120572
(42)
where 119862119871as the definition of (5)
From (42) we obtain that a 100(1 minus 120572) one-sidedcredible interval for 119862
119871is
119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times [Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
(43)
where 119862119871BG is given by (41)
Therefore the 100(1 minus 120572) lower credible interval boundfor 119862119871can be written as
LBBG = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times [Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
(44)
where 119862119871BG is given by (41) 120572 is the specified significance
level and 119887 is a parameter of prior distribution with densityas (15)
The managers can use the one-sided credible interval todetermine whether the product performance attains to therequired level The proposed testing procedure of 119862
119871with
119862119871BG can be organized as follows
Step 1 Determine the lower lifetime limit 119871 for productsand the performance index value 119888
0 then the testing null
hypothesis 1198670 119862119871le 1198880and the alternative hypothesis
1198671 119862119871gt 1198880are constructed
Step 2 Specify a significance level 120572
Step 3 Given the parameters 119886 and 119887 of prior distributionthe parameter 119902 of GE loss function the upper record values119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899)) the lower lifetime limit 119871 and the
significance level 120572 then we can calculate the 100(1 minus 120572)one-sided credible interval [LBBGinfin) for 119862119871 where LBBG asthe definition of (44)
Step 4 The decision rule of statistical test is provided asfollows
(1) if the performance index value 1198880notin [LBBGinfin) then
we will reject 1198670 It is concluded that the lifetime
performance index of products meets the requiredlevel
(2) if the performance index value 1198880isin [LBBGinfin) then
we do not reject 1198670 It is concluded that the lifetime
performance index of products does not meet therequired level
6 The Monte Carlo Simulation Algorithm ofConfidence (or Credible) Level
In this section we will report the results of a simulationstudy for confidence (or credible) level (1 minus 120572) based on a100(1 minus 120572) one-sided confidence (or credible) interval ofthe lifetime performance index 119862
119871 We considered 120572 = 005
and then generated samples from the Rayleigh distributionwith pdf as in (2) with respect to the record values
(i) The Monte Carlo simulation algorithm of confidencelevel (1 minus 120572) for 119862
119871under MLE is given in the
following steps
Step 1 Given 119899 119886 119887 119871 and 120572 where 119886 gt 0 119887 gt 0
Step 2 For the values of prior parameters (119886 119887) use (15) togenerate 120579 from the square-root inverted-gamma distribu-tion
Step 3(a) The generation of data (119885
1 1198852 119885
119899) is by standard
exponential distribution with parameter 1(b) Set1198842
1= 212057921198851and1198842119894= 21205792119885119894+1198842
119894minus1 for 119894 = 2 119899
1198841= radic21205792119885
1 119884
119894= radic21205792119885
119894+ 1198842
119894minus1 for 119894 = 2 119899
(45)
(1198841 1198842 119884
119899) are the upper record values from the
Rayleigh distribution(c) The value of LBMLE is calculated by
LBMLE = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871MLE)(
1205942
21198991minus120572
2119899)
12
(46)
Journal of Applied Mathematics 9
where 119862119871MLE is given by (11) and 1205942
21198991minus120572function
which represents the lower 1 minus 120572 percentile of 12059422119899
(d) If 119862119871ge LBMLE then count = 1 else count = 0
Step 4
(a) Step 3 is repeated 1000 times(b) The estimation of confidence level (1 minus 120572) is (1 minus 120572) =
(total count)1000
Step 5
(a) Repeat Step 2ndashStep 4 for 100 times then we can getthe 100 estimations of confidence level as follows
(1 minus 120572)1 (1 minus 120572)
2 (1 minus 120572)
100 (47)
(b) The average empirical confidence level 1 minus 120572
of (1 minus 120572)119894 119894 = 1 100 that is 1 minus 120572 =
(1100)sum100
119894=1(1 minus 120572)
119894
(c) The sample mean square error (SMSE) of(1 minus 120572)
1 (1 minus 120572)
2 (1 minus 120572)
100SMSE = (1100)
times sum100
119894=1[(1 minus 120572)
119894minus (1 minus 120572)]
2
The results of simulation are summarized in Table 2 basedon 119871 = 10 the different value of sample size 119899 priorparameter (119886 119887) = (2 5) (6 15) and (2 2) at 120572 = 005respectively The scope of SMSE is between 000413 and000593
Step 1 Given 119899 119886 119887 119871 and 120572 where 119886 gt 0 119887 gt 0
Step 2 For the values of prior parameters (119886 119887) use (15) togenerate 120579 from the square-root inverted-gamma distribu-tion
Step 3
(a) The generation of data (1198851 1198852 119885
119899) is by standard
exponential distribution with parameter 1(b) Set1198842
1= 212057921198851and1198842119894= 21205792119885119894+1198842
119894minus1 for 119894 = 2 119899
1198841= radic21205792119885
1 119884
119894= radic21205792119885
119894+ 1198842
119894minus1 for 119894 = 2 119899
(48)
(1198841 1198842 119884
119899) are the upper record values from the
Rayleigh distribution(c) The values of LBBS LBBL and LBBG are calculated by
(25) (36) and (44) respectively(d) If119862
119871ge LB then count = 1 else count = 0 where LB =
LBBS or LBBL or LBBG
Step 4
(a) Step 3 is repeated 1000 times(b) The estimation of credible level (1 minus 120572) is (1 minus 120572) =
(total count)1000
Step 5
(a) Repeat Step 2ndashStep 4 for 100 times then we can getthe 100 estimations of credible level as follows
(1 minus 120572)1 (1 minus 120572)
2 (1 minus 120572)
100 (49)
(b) The average empirical credible level 1 minus 120572 of (1 minus 120572)119894
119894 = 1 100 that is 1 minus 120572 = (1100)sum100119894=1(1 minus 120572)
119894
(c) The sample mean square error (SMSE) of(1 minus 120572)
1 (1 minus 120572)
2 (1 minus 120572)
100SMSE = (1100)
times sum100
119894=1[(1 minus 120572)
119894minus (1 minus 120572)]
2
Based on 119871 = 10 the different value of sample size119899 prior parameter (119886 119887) = (2 5) (6 15) (2 2) and 120572 =
005 the results of simulation are summarized in Tables 3ndash5under SE loss function LINEX loss function with parameter119888 = minus05 05 15 and GE loss function with parameter119902 = 3 5 8 respectively The following points can be drawn
(a) All of the SMSEs are small enough and the scope ofSMSE is between 000346 and 000467
(b) Fix the size 119899 comparison of prior parameter (119886 119887) =(2 5) (6 15) and (2 2) as follows
(i) the values of SMSEwith prior parameter (119886 119887) =(6 15) and (2 5) are smaller than (119886 119887) = (2 2)respectively
(ii) a comparison of prior parameter (119886 119887) = (2 5)and (2 2) that is fix prior parameter 119886 if priorparameter 119887 increases then it can be seen thatthe SMSE will decrease
(c) In Table 4 fix the size 119899 and the prior parameter (119886 119887)comparison the parameter of LINEX loss function 119888 =minus05 05 15 as followsthe values of SMSE with prior parameter 119888 = 05 15andminus15 are the same as the values of SMSE inTable 3
(d) In Table 5 fix the prior parameter (119886 119887) comparisonof GE loss function parameter 119902 = 3 5 8 as followsthe values of SMSE with prior parameter 119902 = 3 5 and8 are the same as the values of SMSE in Table 3
Hence these results from simulation studies illustrate thatthe performance of our proposed method is acceptableMoreover we suggest that the prior parameter (119886 119887) = (6 15)and (2 5) is appropriate for the square-root inverted-gammadistribution
7 Numerical Example
In this section we propose the new hypothesis testingprocedures to a real-life data In the following examplewe discuss a real-life data for 25 ball bearings to illustratethe use of the new hypothesis testing procedures in thelifetime performance of ball bearings The proposed testingprocedures not only can handle nonnormal lifetime data butalso can handle the upper record values
10 Journal of Applied Mathematics
Table 2 Average empirical confidence level (1 minus 120572) for 119862119871under MLE when 120572 = 005
119899 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
5 093653 (000462) 093482 (000467) 093577 (000507)10 093643 (000449) 093581 (000413) 093450 (000593)15 093523 (000489) 093557 (000503) 093463 (000516)119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 3 Average empirical credible level (1 minus 120572) for 119862119871under SE loss function when 120572 = 005
119899 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
5 094357 (000386) 093760 (000386) 093925 (000395)10 094244 (000359) 093704 (000359) 093664 (000462)15 094096 (000346) 093619 (000426) 093635 (000467)119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 4 Average empirical credible level (1 minus 120572) for 119862119871under LINEX loss function when 120572 = 005
119899 119888 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
minus05 094357 (000386) 093760 (000386) 093925 (000395)5 05 094357 (000386) 093760 (000386) 093925 (000395)
15 094357 (000386) 093760 (000386) 093925 (000395)minus05 094244 (000359) 093704 (000359) 093664 (000462)
10 05 094244 (000359) 093704 (000359) 093664 (000462)15 094244 (000359) 093704 (000359) 093664 (000462)minus05 094096 (000346) 093619 (000426) 093635 (000467)
15 05 094096 (000346) 093619 (000426) 093635 (000467)15 094096 (000346) 093619 (000426) 093635 (000467)
119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 5 Average empirical credible level (1 minus 120572) for 119862119871under GE loss function when 120572 = 005
119899 119902 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
3 094357 (000386) 093760 (000386) 093925 (000395)5 5 094357 (000386) 093760 (000386) 093925 (000395)
8 094357 (000386) 093760 (000386) 093925 (000395)3 094244 (000359) 093704 (000359) 093664 (000462)
10 5 094244 (000359) 093704 (000359) 093664 (000462)8 094244 (000359) 093704 (000359) 093664 (000462)3 094096 (000346) 093619 (000426) 093635 (000467)
15 5 094096 (000346) 093619 (000426) 093635 (000467)8 094096 (000346) 093619 (000426) 093635 (000467)
119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Example The data is the failure times of 25 ball bearings inendurance testThe 25 observations are the number ofmillionrevolutions before failure for each of the ball bearings Thedata come from Caroni [27] as follows
6780 6780 6780 6864 3300 6864 9864 12804 42122892 4560 5184 5556 17340 4848 1788 9312 54124152 5196 12792 8412 10512 10584 and 6888
Raqab and Madi [28] and Lee [29] indicated that a one-parameter Rayleigh distribution is acceptable for these dataIn addition we also have a way to test the hypothesis thatthe failure data come from the Rayleigh distribution Thetesting hypothesis119867
0 119883 comes from a Rayleigh distribution
versus1198671 119883 does not come from a Rayleigh distribution is
constructed under significance level 120572 = 005From a probability plot of Figure 1 by using the Minitab
Statistical Software we can conclude the operational lifetimesdata of 25 ball bearings from the Rayleigh distribution whichis the Weibull distribution with the shape 2 (also see Lee etal [14]) For informative prior we use the prior information119864(120579) = 120579MLE ≐ 54834 and Var(120579) = 02 giving the priorparameter values as (119886 = 6014 119887 = 1001)
Here if only the upper record values have been observedthese are 119909
119880(119894) 119894 = 1 5(= 119899) = 6780 6864 9864
12804 17340
Journal of Applied Mathematics 11
10010
99
90807060504030
20
10
5
32
1
Xi
Probability plot of XiWeibull-95 CI
()
Shape 2
Scale 8002N 25
AD 0431P -value 057
Figure 1 Probability plot for failure times of 25 ball bearings data
(1) The proposed testing procedure of 119862119871with 119862
119871MLE isstated as follows
Step 1 The record values from the above data are (119909119880(119894) 119894 =
1 2 3 4 5) = (6780 6864 9864 12804 17340)
Step 2 The lower specification limit 119871 is assumed to be 2337The deal with the product managersrsquo concerns regardinglifetime performance and the conforming rate 119875
119903of products
is required to exceed 80 percent Referring to Table 1 119862119871is
required to exceed 090Thus the performance index value isset at 119888
0= 090 The testing hypothesis119867
0 119862119871le 090 versus
1198671 119862119871gt 090 is constructed
Step 3 Specify a significance level 120572 = 005
Step 4 We can calculate that the 95 lower confidenceinterval bound for 119862
119871 where
LBMLE = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871MLE)(
1205942
21198991minus120572
2119899)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 126251254)(
1205942
10095
10)
12
= 103285
(50)So the 95 one-sided confidence interval for 119862
119871is
[LBMLEinfin) = [103285infin)
Step 5 Because of the performance index value 1198880= 090 notin
[LBMLEinfin) we reject the null hypothesis1198670 119862119871 le 090
Thus we can conclude that the lifetime performanceindex of data meets the required level
(2) The proposed testing procedures of 119862119871with 119862
119871BS119862119871BL and 119862119871BG are stated as follows
Step 1 The record values from the above data are (119909119880(119894) 119894 =
1 2 3 4 5) = (6780 6864 9864 12804 17340)
Step 2 The lower specification limit 119871 is assumed to be 2337The deal with the product managersrsquo concerns regardinglifetime performance and the conforming rate 119875
119903of products
is required to exceed 80 percent Referring to Table 1 119862119871is
required to exceed 090Thus the performance index value isset at 119888
0= 090 The testing hypothesis119867
0 119862119871le 090 versus
1198671 119862119871gt 090 is constructed
Step 3 Specify a significance level 120572 = 005 the parameter ofLINEX loss function 119888 = 05 and the parameter of GE lossfunction 119902 = 2
Step 4 We can calculate the 95 lower credible intervalbound for 119862
119871 where
LBBS = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BS)
timesΓ (119899 + 119887 minus 12)
Γ (119899 + 119887)(1205942
2(119899+119887)1minus120572
2)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1246074685)
timesΓ (5 + 1001 minus 12)
Γ (5 + 1001)(1205942
2(5+1001)095
2)
12
= 096984
(51)
12 Journal of Applied Mathematics
So the 95 one-sided credible interval for 119862119871
is[LBBSinfin) = [096984infin)
LBBL = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1129561059)
times [1
2 (05)(1 minus 119890
minus05(5+1001+1)) 1205942
2(5+1001)095]
12
= 096984
(52)
So the 95 one-sided credible interval for 119862119871
is[LBBLinfin) = [096984infin)
LBBG = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times [Γ (119899 + 119887 + 1199022)
Τ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1200432998)
times [Γ (5 + 1001 + 22)
Γ (5 + 1001)]
minus12
(1205942
2(5+1001)095
2)
12
= 096984
(53)
So the 95 one-sided credible interval for 119862119871
is[LBBGinfin) = [096984infin)
Step 5 Because of the performance index value 1198880= 090 notin
[LBBSinfin) = [LBBLinfin) = [LBBGinfin) we reject the nullhypothesis119867
0 119862119871le 090
Thus we can conclude that the lifetime performance index ofdata meets the required level In addition by using the priorparameter values (119886 119887) = (6 15) based on the simulationstudies we also obtain that 119888
0= 090 notin [LBBSinfin) =
[LBBLinfin) = [LBBGinfin) = 094033 So we also rejectthe null hypothesis 119867
0 119862119871le 090 Hence the lifetime
performance index of data meets the required level
8 Conclusions
Montgomery [1] proposed a process capability index 119862119871for
larger-the-better quality characteristic The assumption ofmost process capability is normal distribution but it is ofteninvalid In this paper we consider that Rayleigh distributionis the special case of Weibull distribution
Moreover in lifetime data testing experiments the exper-imenter may not always be in a position to observe the life
times of all the products put to test In industry and reliabilitystudies many products fail under stress for example anelectronic component ceases to function in an environmentof too high temperature and a battery dies under the stressof time So in such experiments measurement may be madesequentially and only the record values are observed
In order to let the process capability indices can beeffectively used This study constructs MLE and Bayesianestimator of 119862
119871under assuming the conjugate prior distribu-
tion and SE loss function LINEX loss function and GE lossfunction based on the upper record values from the Rayleighdistribution The Bayesian estimator of 119862
119871is then utilized to
develop a credible interval in the condition of known 119871This study also provides a table of the lifetime perfor-
mance index with its corresponding conforming rate basedon the Rayleigh distributionThat is the conforming rate andthe corresponding 119862
119871can be obtained
Further the Bayesian estimator and posterior distributionof 119862119871are also utilized to construct the testing procedure of
119862119871which is based on a credible interval If you want to test
whether the products meet the required level you can utilizethe proposed testing procedure which is easily applied andan effectively method to test in the condition of known 119871Numerical example is illustrated to show that the proposedtesting procedure is effectively evaluating whether the trueperformance index meets requirements
In addition these results from simulation studies illus-trate that the performance of our proposedmethod is accept-able According to SMSE the Bayesian approach is smallerthan the non-Bayesian approach we suggest that the Bayesianapproach is better than the non-Bayesian approachThen theSMSEs of prior parameters (119886 119887) = (6 15) and (2 5) aresmaller than the SMSE of prior parameter (119886 119887) = (2 2)We suggest that the prior parameters (119886 119887) = (6 15) and(2 5) are appropriate for the square-root inverted-gammadistribution based on the Bayesian estimators under theRayleigh distribution
Acknowledgments
The authors thank the referees for their suggestions andhelpful comments on revising the paper This research waspartially supported by the National Science Council China(Plans nos NSC 101-2221-E-158-003 NSC 99-2118-M-415-001 and NSC 100-2118-M-415-002) Moreover J W Wu andC W Hong have a direct financial relation with the NationalScience Council Taiwan but other authors do not a directfinancial relation with the National Science Council Taiwan
References
[1] D C Montgomery Introduction To Statistical Quality ControlJohn Wiley amp Sons New York NY USA 1985
[2] L I Tong K S Chen and H T Chen ldquoStatistical testingfor assessing the performance of lifetime index of electroniccomponents with exponential distributionrdquo Journal of QualityReliability Management vol 19 pp 812ndash824 2002
[3] C-W Hong J-W Wu and C-H Cheng ldquoComputationalprocedure of performance assessment of lifetime index of
Journal of Applied Mathematics 13
businesses for the Pareto lifetime model with the right type IIcensored samplerdquo Applied Mathematics and Computation vol184 no 2 pp 336ndash350 2007
[4] C W Hong J W Wu and C H Cheng ldquoComputational pro-cedure of performance assessment of lifetime index of Paretolifetime businesses based on confidence intervalrdquo Applied SoftComputing vol 8 no 1 pp 698ndash705 2008
[5] J-W Wu H-M Lee and C-L Lei ldquoComputational testingalgorithmic procedure of assessment for lifetime performanceindex of products with two-parameter exponential distribu-tionrdquo Applied Mathematics and Computation vol 190 no 1 pp116ndash125 2007
[6] W-C Lee J-W Wu and C-W Hong ldquoAssessing the lifetimeperformance index of products with the exponential distribu-tion under progressively type II right censored samplesrdquo Journalof Computational and Applied Mathematics vol 231 no 2 pp648ndash656 2009
[7] W C Lee J W Wu and C L Lei ldquoEvaluating the lifetimeperformance index for the exponential lifetime productsrdquoApplied Mathematical Modelling vol 34 no 5 pp 1217ndash12242010
[8] K N Chandler ldquoThe distribution and frequency of recordvaluesrdquo Journal of the Royal Statistical Society B vol 14 pp 220ndash228 1952
[9] S K Bhattacharya and R K Tyagi ldquoBayesian survival analysisbased on the rayleigh modelrdquo Trabajos de Estadistica vol 5 no1 pp 81ndash92 1990
[10] A D Cliff and J K Ord ldquoModel building and the analysis ofspatial pattern in human geography (with discussion)rdquo Journalof the Royal Statistical Society B vol 37 no 3 pp 297ndash348 1975
[11] D D Dyer and C W Whisenand ldquoBest linear unbiasedestimator of the parameter of the Rayleigh distribution I Smallsample theory for censored order statisticsrdquo IEEE Transactionson Reliability vol R-22 no 1 pp 27ndash34 1973
[12] A A Soliman and F M Al-Aboud ldquoBayesian inference usingrecord values from Rayleigh model with applicationrdquo EuropeanJournal of Operational Research vol 185 no 2 pp 659ndash6722008
[13] StatSoft STATISTICA Software Version 7 StatSoft Tulsa Okla2005
[14] W-C Lee J-W Wu M-L Hong L-S Lin and R-L ChanldquoAssessing the lifetime performance index of Rayleigh productsbased on the Bayesian estimation under progressive type IIright censored samplesrdquo Journal of Computational and AppliedMathematics vol 235 no 6 pp 1676ndash1688 2011
[15] M Ahsanullah Record Statistics Nova Science New York NYUSA 1995
[16] B C Arnold N Balakrishnan and H N Nagaraja RecordsJohn Wiley amp Sons New York NY USA 1998
[17] P W Zehna ldquoInvariance of maximum likelihood estimatorsrdquoAnnals of Mathematical Statistics vol 37 p 744 1966
[18] A Zellner ldquoBayesian estimation and prediction using asymmet-ric loss functionsrdquo Journal of the American Statistical Associa-tion vol 81 no 394 pp 446ndash451 1986
[19] S-Y Huang ldquoEmpirical Bayes testing procedures in somenonexponential families using asymmetric linex loss functionrdquoJournal of Statistical Planning and Inference vol 46 no 3 pp293ndash309 1995
[20] R Calabria andG Pulcini ldquoPoint estimation under asymmetricloss functions for left-truncated exponential samplesrdquo Commu-nications in StatisticsmdashTheory and Methods vol 25 no 3 pp585ndash600 1996
[21] A T K Wan G Zou and A H Lee ldquoMinimax and Γ-minimaxestimation for the Poisson distribution under LINEX loss whenthe parameter space is restrictedrdquo Statistics and ProbabilityLetters vol 50 no 1 pp 23ndash32 2000
[22] S Anatolyev ldquoKernel estimation under linear-exponential lossrdquoEconomics Letters vol 91 no 1 pp 39ndash43 2006
[23] X Li Y Shi J Wei and J Chai ldquoEmpirical Bayes estimators ofreliability performances using LINEX loss under progressivelytype-II censored samplesrdquo Mathematics and Computers inSimulation vol 73 no 5 pp 320ndash326 2007
[24] D K Dyer and P S L Liu ldquoOn comparison of estimators in ageneralized life modelrdquo Microelectronics Reliability vol 32 no1-2 pp 207ndash221 1992
[25] A A Soliman ldquoEstimation of parameters of life from progres-sively censored data using Burr-XII modelrdquo IEEE Transactionson Reliability vol 54 no 1 pp 34ndash42 2005
[26] A A Soliman and G R Elkahlout ldquoBayes estimation of thelogistic distribution based on progressively censored samplesrdquoJournal of Applied Statistical Science vol 14 no 3-4 pp 281ndash2932005
[27] C Caroni ldquoThe correct ldquoball bearingsrdquo datardquo Lifetime DataAnalysis vol 8 no 4 pp 395ndash399 2002
[28] M Z Raqab and M T Madi ldquoBayesian prediction of the totaltime on test using doubly censored Rayleigh datardquo Journal ofStatistical Computation and Simulation vol 72 no 10 pp 781ndash789 2002
[29] W C Lee ldquoStatistical testing for assessing lifetime performanceIndex of the Rayleigh lifetime productsrdquo Journal of the ChineseInstitute of Industrial Engineers vol 25 pp 433ndash445 2008
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
6 Journal of Applied Mathematics
where 119862119871BS is given by (22) and 119887 is a parameter of prior
distribution with density as (15)The managers can use the one-sided credible interval to
determine whether the product performance adheres to therequired level The proposed testing procedure of 119862
119871with
119862119871BS can be organized as follows
Step 1 Determine the lower lifetime limit 119871 for productsand the performance index value 119888
0 then the testing null
hypothesis 1198670 119862119871le 1198880and the alternative hypothesis
1198671 119862119871gt 1198880are constructed
Step 2 Specify a significance level 120572
Step 3 Given the parameters 119886 and 119887 of prior distributionthe upper record values 119909
sim= (119909
119880(1) 119909119880(2) 119909
119880(119899)) the
lower lifetime limit 119871 and the significance level 120572 then wecan calculate the 100(1 minus 120572) one-sided credible interval[LBBSinfin) for 119862119871 where LBBS as the definition of (25)
Step 4 The decision rule of statistical test is provided asfollows
(1) if the performance index value 1198880notin [LBBSinfin) then
we will reject 1198670 It is concluded that the lifetime
performance index of products meets the requiredlevel
(2) if the performance index value 1198880isin [LBBSinfin) then
we do not reject 1198670 It is concluded that the lifetime
performance index of products does not meet therequired level
52 Testing Procedure for the Lifetime Performance Index 119862119871
with the Bayesian Estimator under Linear Exponential LossFunction Let 119883 be the lifetime of such a product and 119883has a Rayleigh distribution with the pdf as given by (2) Weconsider the linear exponential (LINEX) loss function as (alsosee [18ndash23])
119871 (Δ) prop 119890119888Δminus 119888Δ minus 1 (26)
where Δ = (120579lowast minus 120579) 119888 = 0In this paper we suppose that
Δ1= (
120579lowast
120579)
2
minus 1 (27)
where 120579lowast is an estimator of 120579By using (26)-(27) the posterior expectation of LINEX
loss function 119871(Δ1) is
119864 [119871 (Δ1) | 119909sim] = 119890minus119888119864 [119890119888(120579lowast
120579)2 | 119909sim]
minus 119888119864 [(120579lowast
120579)
2
minus 1 | 119909sim] minus 1
(28)
where 119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899))
The value of 120579lowast that minimizes 119864[119871(Δ1) | 119909sim] denoted by
120579BL is obtained by solving the equation
119889119864 [119871 (Δ1) | 119909sim]
119889120579lowast= 119890minus119888119864[119890119888(120579lowast
120579)2 sdot2119888120579lowast
1205792| 119909sim]
minus 119888119864(2120579lowast
1205792| 119909sim) = 0
(29)
that is 120579BL is the solution to the following equation
119864[120579BL1205792119890119888(120579BL120579)2 | 119909
sim] = 119890119888119864(
120579BL1205792
| 119909sim) (30)
By using (18) and (30) we have
2120579BL (119899 + 119887) sdot119905119899+119887
(119905 minus 21198881205792
BL)119899+119887+1
= 119890119888120579BL
2 (119899 + 119887)
119905
where 119905 = 1199092119880(119899)
+ 119886
(31)
Hence the Bayesian estimator of 120579 under the LINEX lossfunction is given by
120579BL = [119879
2119888(1 minus 119890
minus119888(119899+119887+1))]
12
(32)
where 119879 = 1198832
119880(119899)+ 119886 119886 and 119887 are the parameters of prior
distribution with density as (15)By using (5) and the Bayesian estimator 120579BL as (32) 119862119871BL
based on the Bayesian estimator 120579BL of 120579 is given by
119862119871BL = radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579BL
= radic120587
4 minus 120587minus radic
2
4 minus 120587119871[119879
2119888(1 minus 119890
minus119888(119899+119887+1))]
minus12
(33)
where 119879 = 1198832
119880(119899)+ 119886 119886 and 119887 are the parameters of prior
distribution with density as (15)We construct a statistical testing procedure to assess
whether the lifetime performance index adheres to therequired level The one-sided credible confidence interval for119862119871is obtained using the pivotal quantity 1199051205792|
119909sim
=(119909119880(1)119909119880(119899))
By using the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899)) given the
specified significance level 120572 the level 100(1minus120572) one-sidedcredible interval for 119862
119871can be derived as follows
Journal of Applied Mathematics 7
Since the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899))sim 1205942
2(119899+119887)
and 12059422(119899+119887)1minus120572
function which represents the lower 1 minus 120572percentile of 1205942
2(119899+119887) then
119875(119905
1205792le 1205942
2(119899+119887)1minus120572| 119909sim) = 1 minus 120572
where 119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899))
997904rArr 119875(119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
| 119909sim)
= 1 minus 120572
(34)
where 119862119871as the definition of (5)
From (34) we obtain that a 100(1minus120572)one-sided credibleinterval for 119862
119871is
119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
(35)
where 119862119871BL is given by (33)
Therefore the 100(1 minus 120572) lower credible interval boundfor 119862119871can be written as
LBBL = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
(36)
where 119862119871BL is given by (33) and 119887 is a parameter of prior
distribution with density as (15)The managers can use the one-sided credible interval to
determine whether the product performance attains to therequired level The proposed testing procedure of 119862
119871with
119862119871BL can be organized as follows
Step 1 Determine the lower lifetime limit 119871 for productsand the performance index value 119888
0 then the testing null
hypothesis 1198670 119862119871le 1198880and the alternative hypothesis
1198671 119862119871gt 1198880are constructed
Step 2 Specify a significance level 120572
Step 3 Given the parameters 119886 and 119887 of prior distribution theparameter 119888 of LINEX loss function the upper record values119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899)) the lower lifetime limit 119871 and the
significance level 120572 then we can calculate the 100(1 minus 120572)one-sided credible interval [LBBLinfin) for 119862119871 where LBBL asthe definition of (36)
Step 4 The decision rule of statistical test is provided asfollows
(1) if the performance index value 1198880notin [LBBLinfin) then
we will reject 1198670 It is concluded that the lifetime
performance index of products meets the requiredlevel
(2) if the performance index value 1198880isin [LBBLinfin) then
we do not reject 1198670 It is concluded that the lifetime
performance index of products does not meet therequired level
53 Testing Procedure for the Lifetime Performance Index119862119871with the Bayesian Estimator under General Entropy Loss
Function Let 119883 be the lifetime of such a product and 119883has a Rayleigh distribution with the pdf as given by (2)We consider the general entropy (GE) loss function (also see[12 20])
119871 (120579lowast 120579) prop (
120579lowast
120579)
119902
minus 119902 log(120579lowast
120579) minus 1 (37)
whose minimum occurs at 120579lowast = 120579The loss function is a generalization of entropy loss used
by several authors (eg Dyer and Liu [24] Soliman [25] andSoliman and Elkahlout [26]) where the shape parameter 119902 =1 When 119902 gt 0 a positive error (120579lowast gt 120579) causes more seriousconsequences than a negative error The Bayesian estimator120579BG of 120579 under GE loss function is given by
120579BG = [119864 (120579minus119902| 119909119880(1) 119909
119880(119899))]minus1119902
(38)
By using (18) and (38) then the Bayesian estimator 120579BG of120579 is derived as follows
119864 (120579minus119902| 119909119880(1) 119909
119880(119899))
= int
infin
0
120579minus1199021199011(120579 | 119909
119880(1) 119909
119880(119899)) 119889120579
= (119905
2)
minus1199022
sdotΓ (119899 + 119887 + 1199022)
Γ (119899 + 119887)
(39)
where 119905 = 1199092119880(119899)
+ 119886 Hence the Bayesian estimator 120579BG of 120579under the GE loss function is given by
120579BG = (119879
2)
12
[Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(40)
where 119879 = 1198832
119880(119899)+ 119886 119886 and 119887 are the parameters of prior
distribution with density as (15)By using (5) and the Bayesian estimator 120579BG as (40)119862
119871BGbased on the Bayesian estimator 120579BG of 120579 is given by
119862119871BG = radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579BG
= radic120587
4 minus 120587minus radic
2
4 minus 120587sdot 119871
sdot [(119879
2)
minus1199022Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
1119902
(41)
8 Journal of Applied Mathematics
where 119879 = 1198832
119880(119899)+ 119886 119886 and 119887 are the parameters of prior
distribution with density as (15)We construct a statistical testing procedure to assess
whether the lifetime performance index adheres to therequired level The one-sided credible confidence interval for119862119871is obtained using the pivotal quantity 1199051205792|
119909sim
=(119909119880(1)119909119880(119899))
By using the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899)) given the
specified significance level 120572 the level 100(1minus120572) one-sidedcredible interval for 119862
119871can be derived as follows
Since the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899))sim 1205942
2(119899+119887)
and 12059422(119899+119887)1minus120572
function which represents the lower 1 minus 120572percentile of 1205942
2(119899+119887) then
119875(119905
1205792le 1205942
2(119899+119887)1minus120572| 119909sim) = 1 minus 120572
where 119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899))
997904rArr 119875(119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times[Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
| 119909sim)
= 1 minus 120572
(42)
where 119862119871as the definition of (5)
From (42) we obtain that a 100(1 minus 120572) one-sidedcredible interval for 119862
119871is
119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times [Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
(43)
where 119862119871BG is given by (41)
Therefore the 100(1 minus 120572) lower credible interval boundfor 119862119871can be written as
LBBG = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times [Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
(44)
where 119862119871BG is given by (41) 120572 is the specified significance
level and 119887 is a parameter of prior distribution with densityas (15)
The managers can use the one-sided credible interval todetermine whether the product performance attains to therequired level The proposed testing procedure of 119862
119871with
119862119871BG can be organized as follows
Step 1 Determine the lower lifetime limit 119871 for productsand the performance index value 119888
0 then the testing null
hypothesis 1198670 119862119871le 1198880and the alternative hypothesis
1198671 119862119871gt 1198880are constructed
Step 2 Specify a significance level 120572
Step 3 Given the parameters 119886 and 119887 of prior distributionthe parameter 119902 of GE loss function the upper record values119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899)) the lower lifetime limit 119871 and the
significance level 120572 then we can calculate the 100(1 minus 120572)one-sided credible interval [LBBGinfin) for 119862119871 where LBBG asthe definition of (44)
Step 4 The decision rule of statistical test is provided asfollows
(1) if the performance index value 1198880notin [LBBGinfin) then
we will reject 1198670 It is concluded that the lifetime
performance index of products meets the requiredlevel
(2) if the performance index value 1198880isin [LBBGinfin) then
we do not reject 1198670 It is concluded that the lifetime
performance index of products does not meet therequired level
6 The Monte Carlo Simulation Algorithm ofConfidence (or Credible) Level
In this section we will report the results of a simulationstudy for confidence (or credible) level (1 minus 120572) based on a100(1 minus 120572) one-sided confidence (or credible) interval ofthe lifetime performance index 119862
119871 We considered 120572 = 005
and then generated samples from the Rayleigh distributionwith pdf as in (2) with respect to the record values
(i) The Monte Carlo simulation algorithm of confidencelevel (1 minus 120572) for 119862
119871under MLE is given in the
following steps
Step 1 Given 119899 119886 119887 119871 and 120572 where 119886 gt 0 119887 gt 0
Step 2 For the values of prior parameters (119886 119887) use (15) togenerate 120579 from the square-root inverted-gamma distribu-tion
Step 3(a) The generation of data (119885
1 1198852 119885
119899) is by standard
exponential distribution with parameter 1(b) Set1198842
1= 212057921198851and1198842119894= 21205792119885119894+1198842
119894minus1 for 119894 = 2 119899
1198841= radic21205792119885
1 119884
119894= radic21205792119885
119894+ 1198842
119894minus1 for 119894 = 2 119899
(45)
(1198841 1198842 119884
119899) are the upper record values from the
Rayleigh distribution(c) The value of LBMLE is calculated by
LBMLE = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871MLE)(
1205942
21198991minus120572
2119899)
12
(46)
Journal of Applied Mathematics 9
where 119862119871MLE is given by (11) and 1205942
21198991minus120572function
which represents the lower 1 minus 120572 percentile of 12059422119899
(d) If 119862119871ge LBMLE then count = 1 else count = 0
Step 4
(a) Step 3 is repeated 1000 times(b) The estimation of confidence level (1 minus 120572) is (1 minus 120572) =
(total count)1000
Step 5
(a) Repeat Step 2ndashStep 4 for 100 times then we can getthe 100 estimations of confidence level as follows
(1 minus 120572)1 (1 minus 120572)
2 (1 minus 120572)
100 (47)
(b) The average empirical confidence level 1 minus 120572
of (1 minus 120572)119894 119894 = 1 100 that is 1 minus 120572 =
(1100)sum100
119894=1(1 minus 120572)
119894
(c) The sample mean square error (SMSE) of(1 minus 120572)
1 (1 minus 120572)
2 (1 minus 120572)
100SMSE = (1100)
times sum100
119894=1[(1 minus 120572)
119894minus (1 minus 120572)]
2
The results of simulation are summarized in Table 2 basedon 119871 = 10 the different value of sample size 119899 priorparameter (119886 119887) = (2 5) (6 15) and (2 2) at 120572 = 005respectively The scope of SMSE is between 000413 and000593
Step 1 Given 119899 119886 119887 119871 and 120572 where 119886 gt 0 119887 gt 0
Step 2 For the values of prior parameters (119886 119887) use (15) togenerate 120579 from the square-root inverted-gamma distribu-tion
Step 3
(a) The generation of data (1198851 1198852 119885
119899) is by standard
exponential distribution with parameter 1(b) Set1198842
1= 212057921198851and1198842119894= 21205792119885119894+1198842
119894minus1 for 119894 = 2 119899
1198841= radic21205792119885
1 119884
119894= radic21205792119885
119894+ 1198842
119894minus1 for 119894 = 2 119899
(48)
(1198841 1198842 119884
119899) are the upper record values from the
Rayleigh distribution(c) The values of LBBS LBBL and LBBG are calculated by
(25) (36) and (44) respectively(d) If119862
119871ge LB then count = 1 else count = 0 where LB =
LBBS or LBBL or LBBG
Step 4
(a) Step 3 is repeated 1000 times(b) The estimation of credible level (1 minus 120572) is (1 minus 120572) =
(total count)1000
Step 5
(a) Repeat Step 2ndashStep 4 for 100 times then we can getthe 100 estimations of credible level as follows
(1 minus 120572)1 (1 minus 120572)
2 (1 minus 120572)
100 (49)
(b) The average empirical credible level 1 minus 120572 of (1 minus 120572)119894
119894 = 1 100 that is 1 minus 120572 = (1100)sum100119894=1(1 minus 120572)
119894
(c) The sample mean square error (SMSE) of(1 minus 120572)
1 (1 minus 120572)
2 (1 minus 120572)
100SMSE = (1100)
times sum100
119894=1[(1 minus 120572)
119894minus (1 minus 120572)]
2
Based on 119871 = 10 the different value of sample size119899 prior parameter (119886 119887) = (2 5) (6 15) (2 2) and 120572 =
005 the results of simulation are summarized in Tables 3ndash5under SE loss function LINEX loss function with parameter119888 = minus05 05 15 and GE loss function with parameter119902 = 3 5 8 respectively The following points can be drawn
(a) All of the SMSEs are small enough and the scope ofSMSE is between 000346 and 000467
(b) Fix the size 119899 comparison of prior parameter (119886 119887) =(2 5) (6 15) and (2 2) as follows
(i) the values of SMSEwith prior parameter (119886 119887) =(6 15) and (2 5) are smaller than (119886 119887) = (2 2)respectively
(ii) a comparison of prior parameter (119886 119887) = (2 5)and (2 2) that is fix prior parameter 119886 if priorparameter 119887 increases then it can be seen thatthe SMSE will decrease
(c) In Table 4 fix the size 119899 and the prior parameter (119886 119887)comparison the parameter of LINEX loss function 119888 =minus05 05 15 as followsthe values of SMSE with prior parameter 119888 = 05 15andminus15 are the same as the values of SMSE inTable 3
(d) In Table 5 fix the prior parameter (119886 119887) comparisonof GE loss function parameter 119902 = 3 5 8 as followsthe values of SMSE with prior parameter 119902 = 3 5 and8 are the same as the values of SMSE in Table 3
Hence these results from simulation studies illustrate thatthe performance of our proposed method is acceptableMoreover we suggest that the prior parameter (119886 119887) = (6 15)and (2 5) is appropriate for the square-root inverted-gammadistribution
7 Numerical Example
In this section we propose the new hypothesis testingprocedures to a real-life data In the following examplewe discuss a real-life data for 25 ball bearings to illustratethe use of the new hypothesis testing procedures in thelifetime performance of ball bearings The proposed testingprocedures not only can handle nonnormal lifetime data butalso can handle the upper record values
10 Journal of Applied Mathematics
Table 2 Average empirical confidence level (1 minus 120572) for 119862119871under MLE when 120572 = 005
119899 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
5 093653 (000462) 093482 (000467) 093577 (000507)10 093643 (000449) 093581 (000413) 093450 (000593)15 093523 (000489) 093557 (000503) 093463 (000516)119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 3 Average empirical credible level (1 minus 120572) for 119862119871under SE loss function when 120572 = 005
119899 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
5 094357 (000386) 093760 (000386) 093925 (000395)10 094244 (000359) 093704 (000359) 093664 (000462)15 094096 (000346) 093619 (000426) 093635 (000467)119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 4 Average empirical credible level (1 minus 120572) for 119862119871under LINEX loss function when 120572 = 005
119899 119888 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
minus05 094357 (000386) 093760 (000386) 093925 (000395)5 05 094357 (000386) 093760 (000386) 093925 (000395)
15 094357 (000386) 093760 (000386) 093925 (000395)minus05 094244 (000359) 093704 (000359) 093664 (000462)
10 05 094244 (000359) 093704 (000359) 093664 (000462)15 094244 (000359) 093704 (000359) 093664 (000462)minus05 094096 (000346) 093619 (000426) 093635 (000467)
15 05 094096 (000346) 093619 (000426) 093635 (000467)15 094096 (000346) 093619 (000426) 093635 (000467)
119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 5 Average empirical credible level (1 minus 120572) for 119862119871under GE loss function when 120572 = 005
119899 119902 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
3 094357 (000386) 093760 (000386) 093925 (000395)5 5 094357 (000386) 093760 (000386) 093925 (000395)
8 094357 (000386) 093760 (000386) 093925 (000395)3 094244 (000359) 093704 (000359) 093664 (000462)
10 5 094244 (000359) 093704 (000359) 093664 (000462)8 094244 (000359) 093704 (000359) 093664 (000462)3 094096 (000346) 093619 (000426) 093635 (000467)
15 5 094096 (000346) 093619 (000426) 093635 (000467)8 094096 (000346) 093619 (000426) 093635 (000467)
119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Example The data is the failure times of 25 ball bearings inendurance testThe 25 observations are the number ofmillionrevolutions before failure for each of the ball bearings Thedata come from Caroni [27] as follows
6780 6780 6780 6864 3300 6864 9864 12804 42122892 4560 5184 5556 17340 4848 1788 9312 54124152 5196 12792 8412 10512 10584 and 6888
Raqab and Madi [28] and Lee [29] indicated that a one-parameter Rayleigh distribution is acceptable for these dataIn addition we also have a way to test the hypothesis thatthe failure data come from the Rayleigh distribution Thetesting hypothesis119867
0 119883 comes from a Rayleigh distribution
versus1198671 119883 does not come from a Rayleigh distribution is
constructed under significance level 120572 = 005From a probability plot of Figure 1 by using the Minitab
Statistical Software we can conclude the operational lifetimesdata of 25 ball bearings from the Rayleigh distribution whichis the Weibull distribution with the shape 2 (also see Lee etal [14]) For informative prior we use the prior information119864(120579) = 120579MLE ≐ 54834 and Var(120579) = 02 giving the priorparameter values as (119886 = 6014 119887 = 1001)
Here if only the upper record values have been observedthese are 119909
119880(119894) 119894 = 1 5(= 119899) = 6780 6864 9864
12804 17340
Journal of Applied Mathematics 11
10010
99
90807060504030
20
10
5
32
1
Xi
Probability plot of XiWeibull-95 CI
()
Shape 2
Scale 8002N 25
AD 0431P -value 057
Figure 1 Probability plot for failure times of 25 ball bearings data
(1) The proposed testing procedure of 119862119871with 119862
119871MLE isstated as follows
Step 1 The record values from the above data are (119909119880(119894) 119894 =
1 2 3 4 5) = (6780 6864 9864 12804 17340)
Step 2 The lower specification limit 119871 is assumed to be 2337The deal with the product managersrsquo concerns regardinglifetime performance and the conforming rate 119875
119903of products
is required to exceed 80 percent Referring to Table 1 119862119871is
required to exceed 090Thus the performance index value isset at 119888
0= 090 The testing hypothesis119867
0 119862119871le 090 versus
1198671 119862119871gt 090 is constructed
Step 3 Specify a significance level 120572 = 005
Step 4 We can calculate that the 95 lower confidenceinterval bound for 119862
119871 where
LBMLE = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871MLE)(
1205942
21198991minus120572
2119899)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 126251254)(
1205942
10095
10)
12
= 103285
(50)So the 95 one-sided confidence interval for 119862
119871is
[LBMLEinfin) = [103285infin)
Step 5 Because of the performance index value 1198880= 090 notin
[LBMLEinfin) we reject the null hypothesis1198670 119862119871 le 090
Thus we can conclude that the lifetime performanceindex of data meets the required level
(2) The proposed testing procedures of 119862119871with 119862
119871BS119862119871BL and 119862119871BG are stated as follows
Step 1 The record values from the above data are (119909119880(119894) 119894 =
1 2 3 4 5) = (6780 6864 9864 12804 17340)
Step 2 The lower specification limit 119871 is assumed to be 2337The deal with the product managersrsquo concerns regardinglifetime performance and the conforming rate 119875
119903of products
is required to exceed 80 percent Referring to Table 1 119862119871is
required to exceed 090Thus the performance index value isset at 119888
0= 090 The testing hypothesis119867
0 119862119871le 090 versus
1198671 119862119871gt 090 is constructed
Step 3 Specify a significance level 120572 = 005 the parameter ofLINEX loss function 119888 = 05 and the parameter of GE lossfunction 119902 = 2
Step 4 We can calculate the 95 lower credible intervalbound for 119862
119871 where
LBBS = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BS)
timesΓ (119899 + 119887 minus 12)
Γ (119899 + 119887)(1205942
2(119899+119887)1minus120572
2)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1246074685)
timesΓ (5 + 1001 minus 12)
Γ (5 + 1001)(1205942
2(5+1001)095
2)
12
= 096984
(51)
12 Journal of Applied Mathematics
So the 95 one-sided credible interval for 119862119871
is[LBBSinfin) = [096984infin)
LBBL = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1129561059)
times [1
2 (05)(1 minus 119890
minus05(5+1001+1)) 1205942
2(5+1001)095]
12
= 096984
(52)
So the 95 one-sided credible interval for 119862119871
is[LBBLinfin) = [096984infin)
LBBG = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times [Γ (119899 + 119887 + 1199022)
Τ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1200432998)
times [Γ (5 + 1001 + 22)
Γ (5 + 1001)]
minus12
(1205942
2(5+1001)095
2)
12
= 096984
(53)
So the 95 one-sided credible interval for 119862119871
is[LBBGinfin) = [096984infin)
Step 5 Because of the performance index value 1198880= 090 notin
[LBBSinfin) = [LBBLinfin) = [LBBGinfin) we reject the nullhypothesis119867
0 119862119871le 090
Thus we can conclude that the lifetime performance index ofdata meets the required level In addition by using the priorparameter values (119886 119887) = (6 15) based on the simulationstudies we also obtain that 119888
0= 090 notin [LBBSinfin) =
[LBBLinfin) = [LBBGinfin) = 094033 So we also rejectthe null hypothesis 119867
0 119862119871le 090 Hence the lifetime
performance index of data meets the required level
8 Conclusions
Montgomery [1] proposed a process capability index 119862119871for
larger-the-better quality characteristic The assumption ofmost process capability is normal distribution but it is ofteninvalid In this paper we consider that Rayleigh distributionis the special case of Weibull distribution
Moreover in lifetime data testing experiments the exper-imenter may not always be in a position to observe the life
times of all the products put to test In industry and reliabilitystudies many products fail under stress for example anelectronic component ceases to function in an environmentof too high temperature and a battery dies under the stressof time So in such experiments measurement may be madesequentially and only the record values are observed
In order to let the process capability indices can beeffectively used This study constructs MLE and Bayesianestimator of 119862
119871under assuming the conjugate prior distribu-
tion and SE loss function LINEX loss function and GE lossfunction based on the upper record values from the Rayleighdistribution The Bayesian estimator of 119862
119871is then utilized to
develop a credible interval in the condition of known 119871This study also provides a table of the lifetime perfor-
mance index with its corresponding conforming rate basedon the Rayleigh distributionThat is the conforming rate andthe corresponding 119862
119871can be obtained
Further the Bayesian estimator and posterior distributionof 119862119871are also utilized to construct the testing procedure of
119862119871which is based on a credible interval If you want to test
whether the products meet the required level you can utilizethe proposed testing procedure which is easily applied andan effectively method to test in the condition of known 119871Numerical example is illustrated to show that the proposedtesting procedure is effectively evaluating whether the trueperformance index meets requirements
In addition these results from simulation studies illus-trate that the performance of our proposedmethod is accept-able According to SMSE the Bayesian approach is smallerthan the non-Bayesian approach we suggest that the Bayesianapproach is better than the non-Bayesian approachThen theSMSEs of prior parameters (119886 119887) = (6 15) and (2 5) aresmaller than the SMSE of prior parameter (119886 119887) = (2 2)We suggest that the prior parameters (119886 119887) = (6 15) and(2 5) are appropriate for the square-root inverted-gammadistribution based on the Bayesian estimators under theRayleigh distribution
Acknowledgments
The authors thank the referees for their suggestions andhelpful comments on revising the paper This research waspartially supported by the National Science Council China(Plans nos NSC 101-2221-E-158-003 NSC 99-2118-M-415-001 and NSC 100-2118-M-415-002) Moreover J W Wu andC W Hong have a direct financial relation with the NationalScience Council Taiwan but other authors do not a directfinancial relation with the National Science Council Taiwan
References
[1] D C Montgomery Introduction To Statistical Quality ControlJohn Wiley amp Sons New York NY USA 1985
[2] L I Tong K S Chen and H T Chen ldquoStatistical testingfor assessing the performance of lifetime index of electroniccomponents with exponential distributionrdquo Journal of QualityReliability Management vol 19 pp 812ndash824 2002
[3] C-W Hong J-W Wu and C-H Cheng ldquoComputationalprocedure of performance assessment of lifetime index of
Journal of Applied Mathematics 13
businesses for the Pareto lifetime model with the right type IIcensored samplerdquo Applied Mathematics and Computation vol184 no 2 pp 336ndash350 2007
[4] C W Hong J W Wu and C H Cheng ldquoComputational pro-cedure of performance assessment of lifetime index of Paretolifetime businesses based on confidence intervalrdquo Applied SoftComputing vol 8 no 1 pp 698ndash705 2008
[5] J-W Wu H-M Lee and C-L Lei ldquoComputational testingalgorithmic procedure of assessment for lifetime performanceindex of products with two-parameter exponential distribu-tionrdquo Applied Mathematics and Computation vol 190 no 1 pp116ndash125 2007
[6] W-C Lee J-W Wu and C-W Hong ldquoAssessing the lifetimeperformance index of products with the exponential distribu-tion under progressively type II right censored samplesrdquo Journalof Computational and Applied Mathematics vol 231 no 2 pp648ndash656 2009
[7] W C Lee J W Wu and C L Lei ldquoEvaluating the lifetimeperformance index for the exponential lifetime productsrdquoApplied Mathematical Modelling vol 34 no 5 pp 1217ndash12242010
[8] K N Chandler ldquoThe distribution and frequency of recordvaluesrdquo Journal of the Royal Statistical Society B vol 14 pp 220ndash228 1952
[9] S K Bhattacharya and R K Tyagi ldquoBayesian survival analysisbased on the rayleigh modelrdquo Trabajos de Estadistica vol 5 no1 pp 81ndash92 1990
[10] A D Cliff and J K Ord ldquoModel building and the analysis ofspatial pattern in human geography (with discussion)rdquo Journalof the Royal Statistical Society B vol 37 no 3 pp 297ndash348 1975
[11] D D Dyer and C W Whisenand ldquoBest linear unbiasedestimator of the parameter of the Rayleigh distribution I Smallsample theory for censored order statisticsrdquo IEEE Transactionson Reliability vol R-22 no 1 pp 27ndash34 1973
[12] A A Soliman and F M Al-Aboud ldquoBayesian inference usingrecord values from Rayleigh model with applicationrdquo EuropeanJournal of Operational Research vol 185 no 2 pp 659ndash6722008
[13] StatSoft STATISTICA Software Version 7 StatSoft Tulsa Okla2005
[14] W-C Lee J-W Wu M-L Hong L-S Lin and R-L ChanldquoAssessing the lifetime performance index of Rayleigh productsbased on the Bayesian estimation under progressive type IIright censored samplesrdquo Journal of Computational and AppliedMathematics vol 235 no 6 pp 1676ndash1688 2011
[15] M Ahsanullah Record Statistics Nova Science New York NYUSA 1995
[16] B C Arnold N Balakrishnan and H N Nagaraja RecordsJohn Wiley amp Sons New York NY USA 1998
[17] P W Zehna ldquoInvariance of maximum likelihood estimatorsrdquoAnnals of Mathematical Statistics vol 37 p 744 1966
[18] A Zellner ldquoBayesian estimation and prediction using asymmet-ric loss functionsrdquo Journal of the American Statistical Associa-tion vol 81 no 394 pp 446ndash451 1986
[19] S-Y Huang ldquoEmpirical Bayes testing procedures in somenonexponential families using asymmetric linex loss functionrdquoJournal of Statistical Planning and Inference vol 46 no 3 pp293ndash309 1995
[20] R Calabria andG Pulcini ldquoPoint estimation under asymmetricloss functions for left-truncated exponential samplesrdquo Commu-nications in StatisticsmdashTheory and Methods vol 25 no 3 pp585ndash600 1996
[21] A T K Wan G Zou and A H Lee ldquoMinimax and Γ-minimaxestimation for the Poisson distribution under LINEX loss whenthe parameter space is restrictedrdquo Statistics and ProbabilityLetters vol 50 no 1 pp 23ndash32 2000
[22] S Anatolyev ldquoKernel estimation under linear-exponential lossrdquoEconomics Letters vol 91 no 1 pp 39ndash43 2006
[23] X Li Y Shi J Wei and J Chai ldquoEmpirical Bayes estimators ofreliability performances using LINEX loss under progressivelytype-II censored samplesrdquo Mathematics and Computers inSimulation vol 73 no 5 pp 320ndash326 2007
[24] D K Dyer and P S L Liu ldquoOn comparison of estimators in ageneralized life modelrdquo Microelectronics Reliability vol 32 no1-2 pp 207ndash221 1992
[25] A A Soliman ldquoEstimation of parameters of life from progres-sively censored data using Burr-XII modelrdquo IEEE Transactionson Reliability vol 54 no 1 pp 34ndash42 2005
[26] A A Soliman and G R Elkahlout ldquoBayes estimation of thelogistic distribution based on progressively censored samplesrdquoJournal of Applied Statistical Science vol 14 no 3-4 pp 281ndash2932005
[27] C Caroni ldquoThe correct ldquoball bearingsrdquo datardquo Lifetime DataAnalysis vol 8 no 4 pp 395ndash399 2002
[28] M Z Raqab and M T Madi ldquoBayesian prediction of the totaltime on test using doubly censored Rayleigh datardquo Journal ofStatistical Computation and Simulation vol 72 no 10 pp 781ndash789 2002
[29] W C Lee ldquoStatistical testing for assessing lifetime performanceIndex of the Rayleigh lifetime productsrdquo Journal of the ChineseInstitute of Industrial Engineers vol 25 pp 433ndash445 2008
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Journal of Applied Mathematics 7
Since the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899))sim 1205942
2(119899+119887)
and 12059422(119899+119887)1minus120572
function which represents the lower 1 minus 120572percentile of 1205942
2(119899+119887) then
119875(119905
1205792le 1205942
2(119899+119887)1minus120572| 119909sim) = 1 minus 120572
where 119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899))
997904rArr 119875(119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
| 119909sim)
= 1 minus 120572
(34)
where 119862119871as the definition of (5)
From (34) we obtain that a 100(1minus120572)one-sided credibleinterval for 119862
119871is
119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
(35)
where 119862119871BL is given by (33)
Therefore the 100(1 minus 120572) lower credible interval boundfor 119862119871can be written as
LBBL = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
(36)
where 119862119871BL is given by (33) and 119887 is a parameter of prior
distribution with density as (15)The managers can use the one-sided credible interval to
determine whether the product performance attains to therequired level The proposed testing procedure of 119862
119871with
119862119871BL can be organized as follows
Step 1 Determine the lower lifetime limit 119871 for productsand the performance index value 119888
0 then the testing null
hypothesis 1198670 119862119871le 1198880and the alternative hypothesis
1198671 119862119871gt 1198880are constructed
Step 2 Specify a significance level 120572
Step 3 Given the parameters 119886 and 119887 of prior distribution theparameter 119888 of LINEX loss function the upper record values119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899)) the lower lifetime limit 119871 and the
significance level 120572 then we can calculate the 100(1 minus 120572)one-sided credible interval [LBBLinfin) for 119862119871 where LBBL asthe definition of (36)
Step 4 The decision rule of statistical test is provided asfollows
(1) if the performance index value 1198880notin [LBBLinfin) then
we will reject 1198670 It is concluded that the lifetime
performance index of products meets the requiredlevel
(2) if the performance index value 1198880isin [LBBLinfin) then
we do not reject 1198670 It is concluded that the lifetime
performance index of products does not meet therequired level
53 Testing Procedure for the Lifetime Performance Index119862119871with the Bayesian Estimator under General Entropy Loss
Function Let 119883 be the lifetime of such a product and 119883has a Rayleigh distribution with the pdf as given by (2)We consider the general entropy (GE) loss function (also see[12 20])
119871 (120579lowast 120579) prop (
120579lowast
120579)
119902
minus 119902 log(120579lowast
120579) minus 1 (37)
whose minimum occurs at 120579lowast = 120579The loss function is a generalization of entropy loss used
by several authors (eg Dyer and Liu [24] Soliman [25] andSoliman and Elkahlout [26]) where the shape parameter 119902 =1 When 119902 gt 0 a positive error (120579lowast gt 120579) causes more seriousconsequences than a negative error The Bayesian estimator120579BG of 120579 under GE loss function is given by
120579BG = [119864 (120579minus119902| 119909119880(1) 119909
119880(119899))]minus1119902
(38)
By using (18) and (38) then the Bayesian estimator 120579BG of120579 is derived as follows
119864 (120579minus119902| 119909119880(1) 119909
119880(119899))
= int
infin
0
120579minus1199021199011(120579 | 119909
119880(1) 119909
119880(119899)) 119889120579
= (119905
2)
minus1199022
sdotΓ (119899 + 119887 + 1199022)
Γ (119899 + 119887)
(39)
where 119905 = 1199092119880(119899)
+ 119886 Hence the Bayesian estimator 120579BG of 120579under the GE loss function is given by
120579BG = (119879
2)
12
[Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(40)
where 119879 = 1198832
119880(119899)+ 119886 119886 and 119887 are the parameters of prior
distribution with density as (15)By using (5) and the Bayesian estimator 120579BG as (40)119862
119871BGbased on the Bayesian estimator 120579BG of 120579 is given by
119862119871BG = radic
120587
4 minus 120587minus radic
2
4 minus 120587
119871
120579BG
= radic120587
4 minus 120587minus radic
2
4 minus 120587sdot 119871
sdot [(119879
2)
minus1199022Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
1119902
(41)
8 Journal of Applied Mathematics
where 119879 = 1198832
119880(119899)+ 119886 119886 and 119887 are the parameters of prior
distribution with density as (15)We construct a statistical testing procedure to assess
whether the lifetime performance index adheres to therequired level The one-sided credible confidence interval for119862119871is obtained using the pivotal quantity 1199051205792|
119909sim
=(119909119880(1)119909119880(119899))
By using the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899)) given the
specified significance level 120572 the level 100(1minus120572) one-sidedcredible interval for 119862
119871can be derived as follows
Since the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899))sim 1205942
2(119899+119887)
and 12059422(119899+119887)1minus120572
function which represents the lower 1 minus 120572percentile of 1205942
2(119899+119887) then
119875(119905
1205792le 1205942
2(119899+119887)1minus120572| 119909sim) = 1 minus 120572
where 119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899))
997904rArr 119875(119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times[Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
| 119909sim)
= 1 minus 120572
(42)
where 119862119871as the definition of (5)
From (42) we obtain that a 100(1 minus 120572) one-sidedcredible interval for 119862
119871is
119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times [Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
(43)
where 119862119871BG is given by (41)
Therefore the 100(1 minus 120572) lower credible interval boundfor 119862119871can be written as
LBBG = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times [Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
(44)
where 119862119871BG is given by (41) 120572 is the specified significance
level and 119887 is a parameter of prior distribution with densityas (15)
The managers can use the one-sided credible interval todetermine whether the product performance attains to therequired level The proposed testing procedure of 119862
119871with
119862119871BG can be organized as follows
Step 1 Determine the lower lifetime limit 119871 for productsand the performance index value 119888
0 then the testing null
hypothesis 1198670 119862119871le 1198880and the alternative hypothesis
1198671 119862119871gt 1198880are constructed
Step 2 Specify a significance level 120572
Step 3 Given the parameters 119886 and 119887 of prior distributionthe parameter 119902 of GE loss function the upper record values119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899)) the lower lifetime limit 119871 and the
significance level 120572 then we can calculate the 100(1 minus 120572)one-sided credible interval [LBBGinfin) for 119862119871 where LBBG asthe definition of (44)
Step 4 The decision rule of statistical test is provided asfollows
(1) if the performance index value 1198880notin [LBBGinfin) then
we will reject 1198670 It is concluded that the lifetime
performance index of products meets the requiredlevel
(2) if the performance index value 1198880isin [LBBGinfin) then
we do not reject 1198670 It is concluded that the lifetime
performance index of products does not meet therequired level
6 The Monte Carlo Simulation Algorithm ofConfidence (or Credible) Level
In this section we will report the results of a simulationstudy for confidence (or credible) level (1 minus 120572) based on a100(1 minus 120572) one-sided confidence (or credible) interval ofthe lifetime performance index 119862
119871 We considered 120572 = 005
and then generated samples from the Rayleigh distributionwith pdf as in (2) with respect to the record values
(i) The Monte Carlo simulation algorithm of confidencelevel (1 minus 120572) for 119862
119871under MLE is given in the
following steps
Step 1 Given 119899 119886 119887 119871 and 120572 where 119886 gt 0 119887 gt 0
Step 2 For the values of prior parameters (119886 119887) use (15) togenerate 120579 from the square-root inverted-gamma distribu-tion
Step 3(a) The generation of data (119885
1 1198852 119885
119899) is by standard
exponential distribution with parameter 1(b) Set1198842
1= 212057921198851and1198842119894= 21205792119885119894+1198842
119894minus1 for 119894 = 2 119899
1198841= radic21205792119885
1 119884
119894= radic21205792119885
119894+ 1198842
119894minus1 for 119894 = 2 119899
(45)
(1198841 1198842 119884
119899) are the upper record values from the
Rayleigh distribution(c) The value of LBMLE is calculated by
LBMLE = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871MLE)(
1205942
21198991minus120572
2119899)
12
(46)
Journal of Applied Mathematics 9
where 119862119871MLE is given by (11) and 1205942
21198991minus120572function
which represents the lower 1 minus 120572 percentile of 12059422119899
(d) If 119862119871ge LBMLE then count = 1 else count = 0
Step 4
(a) Step 3 is repeated 1000 times(b) The estimation of confidence level (1 minus 120572) is (1 minus 120572) =
(total count)1000
Step 5
(a) Repeat Step 2ndashStep 4 for 100 times then we can getthe 100 estimations of confidence level as follows
(1 minus 120572)1 (1 minus 120572)
2 (1 minus 120572)
100 (47)
(b) The average empirical confidence level 1 minus 120572
of (1 minus 120572)119894 119894 = 1 100 that is 1 minus 120572 =
(1100)sum100
119894=1(1 minus 120572)
119894
(c) The sample mean square error (SMSE) of(1 minus 120572)
1 (1 minus 120572)
2 (1 minus 120572)
100SMSE = (1100)
times sum100
119894=1[(1 minus 120572)
119894minus (1 minus 120572)]
2
The results of simulation are summarized in Table 2 basedon 119871 = 10 the different value of sample size 119899 priorparameter (119886 119887) = (2 5) (6 15) and (2 2) at 120572 = 005respectively The scope of SMSE is between 000413 and000593
Step 1 Given 119899 119886 119887 119871 and 120572 where 119886 gt 0 119887 gt 0
Step 2 For the values of prior parameters (119886 119887) use (15) togenerate 120579 from the square-root inverted-gamma distribu-tion
Step 3
(a) The generation of data (1198851 1198852 119885
119899) is by standard
exponential distribution with parameter 1(b) Set1198842
1= 212057921198851and1198842119894= 21205792119885119894+1198842
119894minus1 for 119894 = 2 119899
1198841= radic21205792119885
1 119884
119894= radic21205792119885
119894+ 1198842
119894minus1 for 119894 = 2 119899
(48)
(1198841 1198842 119884
119899) are the upper record values from the
Rayleigh distribution(c) The values of LBBS LBBL and LBBG are calculated by
(25) (36) and (44) respectively(d) If119862
119871ge LB then count = 1 else count = 0 where LB =
LBBS or LBBL or LBBG
Step 4
(a) Step 3 is repeated 1000 times(b) The estimation of credible level (1 minus 120572) is (1 minus 120572) =
(total count)1000
Step 5
(a) Repeat Step 2ndashStep 4 for 100 times then we can getthe 100 estimations of credible level as follows
(1 minus 120572)1 (1 minus 120572)
2 (1 minus 120572)
100 (49)
(b) The average empirical credible level 1 minus 120572 of (1 minus 120572)119894
119894 = 1 100 that is 1 minus 120572 = (1100)sum100119894=1(1 minus 120572)
119894
(c) The sample mean square error (SMSE) of(1 minus 120572)
1 (1 minus 120572)
2 (1 minus 120572)
100SMSE = (1100)
times sum100
119894=1[(1 minus 120572)
119894minus (1 minus 120572)]
2
Based on 119871 = 10 the different value of sample size119899 prior parameter (119886 119887) = (2 5) (6 15) (2 2) and 120572 =
005 the results of simulation are summarized in Tables 3ndash5under SE loss function LINEX loss function with parameter119888 = minus05 05 15 and GE loss function with parameter119902 = 3 5 8 respectively The following points can be drawn
(a) All of the SMSEs are small enough and the scope ofSMSE is between 000346 and 000467
(b) Fix the size 119899 comparison of prior parameter (119886 119887) =(2 5) (6 15) and (2 2) as follows
(i) the values of SMSEwith prior parameter (119886 119887) =(6 15) and (2 5) are smaller than (119886 119887) = (2 2)respectively
(ii) a comparison of prior parameter (119886 119887) = (2 5)and (2 2) that is fix prior parameter 119886 if priorparameter 119887 increases then it can be seen thatthe SMSE will decrease
(c) In Table 4 fix the size 119899 and the prior parameter (119886 119887)comparison the parameter of LINEX loss function 119888 =minus05 05 15 as followsthe values of SMSE with prior parameter 119888 = 05 15andminus15 are the same as the values of SMSE inTable 3
(d) In Table 5 fix the prior parameter (119886 119887) comparisonof GE loss function parameter 119902 = 3 5 8 as followsthe values of SMSE with prior parameter 119902 = 3 5 and8 are the same as the values of SMSE in Table 3
Hence these results from simulation studies illustrate thatthe performance of our proposed method is acceptableMoreover we suggest that the prior parameter (119886 119887) = (6 15)and (2 5) is appropriate for the square-root inverted-gammadistribution
7 Numerical Example
In this section we propose the new hypothesis testingprocedures to a real-life data In the following examplewe discuss a real-life data for 25 ball bearings to illustratethe use of the new hypothesis testing procedures in thelifetime performance of ball bearings The proposed testingprocedures not only can handle nonnormal lifetime data butalso can handle the upper record values
10 Journal of Applied Mathematics
Table 2 Average empirical confidence level (1 minus 120572) for 119862119871under MLE when 120572 = 005
119899 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
5 093653 (000462) 093482 (000467) 093577 (000507)10 093643 (000449) 093581 (000413) 093450 (000593)15 093523 (000489) 093557 (000503) 093463 (000516)119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 3 Average empirical credible level (1 minus 120572) for 119862119871under SE loss function when 120572 = 005
119899 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
5 094357 (000386) 093760 (000386) 093925 (000395)10 094244 (000359) 093704 (000359) 093664 (000462)15 094096 (000346) 093619 (000426) 093635 (000467)119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 4 Average empirical credible level (1 minus 120572) for 119862119871under LINEX loss function when 120572 = 005
119899 119888 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
minus05 094357 (000386) 093760 (000386) 093925 (000395)5 05 094357 (000386) 093760 (000386) 093925 (000395)
15 094357 (000386) 093760 (000386) 093925 (000395)minus05 094244 (000359) 093704 (000359) 093664 (000462)
10 05 094244 (000359) 093704 (000359) 093664 (000462)15 094244 (000359) 093704 (000359) 093664 (000462)minus05 094096 (000346) 093619 (000426) 093635 (000467)
15 05 094096 (000346) 093619 (000426) 093635 (000467)15 094096 (000346) 093619 (000426) 093635 (000467)
119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 5 Average empirical credible level (1 minus 120572) for 119862119871under GE loss function when 120572 = 005
119899 119902 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
3 094357 (000386) 093760 (000386) 093925 (000395)5 5 094357 (000386) 093760 (000386) 093925 (000395)
8 094357 (000386) 093760 (000386) 093925 (000395)3 094244 (000359) 093704 (000359) 093664 (000462)
10 5 094244 (000359) 093704 (000359) 093664 (000462)8 094244 (000359) 093704 (000359) 093664 (000462)3 094096 (000346) 093619 (000426) 093635 (000467)
15 5 094096 (000346) 093619 (000426) 093635 (000467)8 094096 (000346) 093619 (000426) 093635 (000467)
119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Example The data is the failure times of 25 ball bearings inendurance testThe 25 observations are the number ofmillionrevolutions before failure for each of the ball bearings Thedata come from Caroni [27] as follows
6780 6780 6780 6864 3300 6864 9864 12804 42122892 4560 5184 5556 17340 4848 1788 9312 54124152 5196 12792 8412 10512 10584 and 6888
Raqab and Madi [28] and Lee [29] indicated that a one-parameter Rayleigh distribution is acceptable for these dataIn addition we also have a way to test the hypothesis thatthe failure data come from the Rayleigh distribution Thetesting hypothesis119867
0 119883 comes from a Rayleigh distribution
versus1198671 119883 does not come from a Rayleigh distribution is
constructed under significance level 120572 = 005From a probability plot of Figure 1 by using the Minitab
Statistical Software we can conclude the operational lifetimesdata of 25 ball bearings from the Rayleigh distribution whichis the Weibull distribution with the shape 2 (also see Lee etal [14]) For informative prior we use the prior information119864(120579) = 120579MLE ≐ 54834 and Var(120579) = 02 giving the priorparameter values as (119886 = 6014 119887 = 1001)
Here if only the upper record values have been observedthese are 119909
119880(119894) 119894 = 1 5(= 119899) = 6780 6864 9864
12804 17340
Journal of Applied Mathematics 11
10010
99
90807060504030
20
10
5
32
1
Xi
Probability plot of XiWeibull-95 CI
()
Shape 2
Scale 8002N 25
AD 0431P -value 057
Figure 1 Probability plot for failure times of 25 ball bearings data
(1) The proposed testing procedure of 119862119871with 119862
119871MLE isstated as follows
Step 1 The record values from the above data are (119909119880(119894) 119894 =
1 2 3 4 5) = (6780 6864 9864 12804 17340)
Step 2 The lower specification limit 119871 is assumed to be 2337The deal with the product managersrsquo concerns regardinglifetime performance and the conforming rate 119875
119903of products
is required to exceed 80 percent Referring to Table 1 119862119871is
required to exceed 090Thus the performance index value isset at 119888
0= 090 The testing hypothesis119867
0 119862119871le 090 versus
1198671 119862119871gt 090 is constructed
Step 3 Specify a significance level 120572 = 005
Step 4 We can calculate that the 95 lower confidenceinterval bound for 119862
119871 where
LBMLE = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871MLE)(
1205942
21198991minus120572
2119899)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 126251254)(
1205942
10095
10)
12
= 103285
(50)So the 95 one-sided confidence interval for 119862
119871is
[LBMLEinfin) = [103285infin)
Step 5 Because of the performance index value 1198880= 090 notin
[LBMLEinfin) we reject the null hypothesis1198670 119862119871 le 090
Thus we can conclude that the lifetime performanceindex of data meets the required level
(2) The proposed testing procedures of 119862119871with 119862
119871BS119862119871BL and 119862119871BG are stated as follows
Step 1 The record values from the above data are (119909119880(119894) 119894 =
1 2 3 4 5) = (6780 6864 9864 12804 17340)
Step 2 The lower specification limit 119871 is assumed to be 2337The deal with the product managersrsquo concerns regardinglifetime performance and the conforming rate 119875
119903of products
is required to exceed 80 percent Referring to Table 1 119862119871is
required to exceed 090Thus the performance index value isset at 119888
0= 090 The testing hypothesis119867
0 119862119871le 090 versus
1198671 119862119871gt 090 is constructed
Step 3 Specify a significance level 120572 = 005 the parameter ofLINEX loss function 119888 = 05 and the parameter of GE lossfunction 119902 = 2
Step 4 We can calculate the 95 lower credible intervalbound for 119862
119871 where
LBBS = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BS)
timesΓ (119899 + 119887 minus 12)
Γ (119899 + 119887)(1205942
2(119899+119887)1minus120572
2)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1246074685)
timesΓ (5 + 1001 minus 12)
Γ (5 + 1001)(1205942
2(5+1001)095
2)
12
= 096984
(51)
12 Journal of Applied Mathematics
So the 95 one-sided credible interval for 119862119871
is[LBBSinfin) = [096984infin)
LBBL = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1129561059)
times [1
2 (05)(1 minus 119890
minus05(5+1001+1)) 1205942
2(5+1001)095]
12
= 096984
(52)
So the 95 one-sided credible interval for 119862119871
is[LBBLinfin) = [096984infin)
LBBG = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times [Γ (119899 + 119887 + 1199022)
Τ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1200432998)
times [Γ (5 + 1001 + 22)
Γ (5 + 1001)]
minus12
(1205942
2(5+1001)095
2)
12
= 096984
(53)
So the 95 one-sided credible interval for 119862119871
is[LBBGinfin) = [096984infin)
Step 5 Because of the performance index value 1198880= 090 notin
[LBBSinfin) = [LBBLinfin) = [LBBGinfin) we reject the nullhypothesis119867
0 119862119871le 090
Thus we can conclude that the lifetime performance index ofdata meets the required level In addition by using the priorparameter values (119886 119887) = (6 15) based on the simulationstudies we also obtain that 119888
0= 090 notin [LBBSinfin) =
[LBBLinfin) = [LBBGinfin) = 094033 So we also rejectthe null hypothesis 119867
0 119862119871le 090 Hence the lifetime
performance index of data meets the required level
8 Conclusions
Montgomery [1] proposed a process capability index 119862119871for
larger-the-better quality characteristic The assumption ofmost process capability is normal distribution but it is ofteninvalid In this paper we consider that Rayleigh distributionis the special case of Weibull distribution
Moreover in lifetime data testing experiments the exper-imenter may not always be in a position to observe the life
times of all the products put to test In industry and reliabilitystudies many products fail under stress for example anelectronic component ceases to function in an environmentof too high temperature and a battery dies under the stressof time So in such experiments measurement may be madesequentially and only the record values are observed
In order to let the process capability indices can beeffectively used This study constructs MLE and Bayesianestimator of 119862
119871under assuming the conjugate prior distribu-
tion and SE loss function LINEX loss function and GE lossfunction based on the upper record values from the Rayleighdistribution The Bayesian estimator of 119862
119871is then utilized to
develop a credible interval in the condition of known 119871This study also provides a table of the lifetime perfor-
mance index with its corresponding conforming rate basedon the Rayleigh distributionThat is the conforming rate andthe corresponding 119862
119871can be obtained
Further the Bayesian estimator and posterior distributionof 119862119871are also utilized to construct the testing procedure of
119862119871which is based on a credible interval If you want to test
whether the products meet the required level you can utilizethe proposed testing procedure which is easily applied andan effectively method to test in the condition of known 119871Numerical example is illustrated to show that the proposedtesting procedure is effectively evaluating whether the trueperformance index meets requirements
In addition these results from simulation studies illus-trate that the performance of our proposedmethod is accept-able According to SMSE the Bayesian approach is smallerthan the non-Bayesian approach we suggest that the Bayesianapproach is better than the non-Bayesian approachThen theSMSEs of prior parameters (119886 119887) = (6 15) and (2 5) aresmaller than the SMSE of prior parameter (119886 119887) = (2 2)We suggest that the prior parameters (119886 119887) = (6 15) and(2 5) are appropriate for the square-root inverted-gammadistribution based on the Bayesian estimators under theRayleigh distribution
Acknowledgments
The authors thank the referees for their suggestions andhelpful comments on revising the paper This research waspartially supported by the National Science Council China(Plans nos NSC 101-2221-E-158-003 NSC 99-2118-M-415-001 and NSC 100-2118-M-415-002) Moreover J W Wu andC W Hong have a direct financial relation with the NationalScience Council Taiwan but other authors do not a directfinancial relation with the National Science Council Taiwan
References
[1] D C Montgomery Introduction To Statistical Quality ControlJohn Wiley amp Sons New York NY USA 1985
[2] L I Tong K S Chen and H T Chen ldquoStatistical testingfor assessing the performance of lifetime index of electroniccomponents with exponential distributionrdquo Journal of QualityReliability Management vol 19 pp 812ndash824 2002
[3] C-W Hong J-W Wu and C-H Cheng ldquoComputationalprocedure of performance assessment of lifetime index of
Journal of Applied Mathematics 13
businesses for the Pareto lifetime model with the right type IIcensored samplerdquo Applied Mathematics and Computation vol184 no 2 pp 336ndash350 2007
[4] C W Hong J W Wu and C H Cheng ldquoComputational pro-cedure of performance assessment of lifetime index of Paretolifetime businesses based on confidence intervalrdquo Applied SoftComputing vol 8 no 1 pp 698ndash705 2008
[5] J-W Wu H-M Lee and C-L Lei ldquoComputational testingalgorithmic procedure of assessment for lifetime performanceindex of products with two-parameter exponential distribu-tionrdquo Applied Mathematics and Computation vol 190 no 1 pp116ndash125 2007
[6] W-C Lee J-W Wu and C-W Hong ldquoAssessing the lifetimeperformance index of products with the exponential distribu-tion under progressively type II right censored samplesrdquo Journalof Computational and Applied Mathematics vol 231 no 2 pp648ndash656 2009
[7] W C Lee J W Wu and C L Lei ldquoEvaluating the lifetimeperformance index for the exponential lifetime productsrdquoApplied Mathematical Modelling vol 34 no 5 pp 1217ndash12242010
[8] K N Chandler ldquoThe distribution and frequency of recordvaluesrdquo Journal of the Royal Statistical Society B vol 14 pp 220ndash228 1952
[9] S K Bhattacharya and R K Tyagi ldquoBayesian survival analysisbased on the rayleigh modelrdquo Trabajos de Estadistica vol 5 no1 pp 81ndash92 1990
[10] A D Cliff and J K Ord ldquoModel building and the analysis ofspatial pattern in human geography (with discussion)rdquo Journalof the Royal Statistical Society B vol 37 no 3 pp 297ndash348 1975
[11] D D Dyer and C W Whisenand ldquoBest linear unbiasedestimator of the parameter of the Rayleigh distribution I Smallsample theory for censored order statisticsrdquo IEEE Transactionson Reliability vol R-22 no 1 pp 27ndash34 1973
[12] A A Soliman and F M Al-Aboud ldquoBayesian inference usingrecord values from Rayleigh model with applicationrdquo EuropeanJournal of Operational Research vol 185 no 2 pp 659ndash6722008
[13] StatSoft STATISTICA Software Version 7 StatSoft Tulsa Okla2005
[14] W-C Lee J-W Wu M-L Hong L-S Lin and R-L ChanldquoAssessing the lifetime performance index of Rayleigh productsbased on the Bayesian estimation under progressive type IIright censored samplesrdquo Journal of Computational and AppliedMathematics vol 235 no 6 pp 1676ndash1688 2011
[15] M Ahsanullah Record Statistics Nova Science New York NYUSA 1995
[16] B C Arnold N Balakrishnan and H N Nagaraja RecordsJohn Wiley amp Sons New York NY USA 1998
[17] P W Zehna ldquoInvariance of maximum likelihood estimatorsrdquoAnnals of Mathematical Statistics vol 37 p 744 1966
[18] A Zellner ldquoBayesian estimation and prediction using asymmet-ric loss functionsrdquo Journal of the American Statistical Associa-tion vol 81 no 394 pp 446ndash451 1986
[19] S-Y Huang ldquoEmpirical Bayes testing procedures in somenonexponential families using asymmetric linex loss functionrdquoJournal of Statistical Planning and Inference vol 46 no 3 pp293ndash309 1995
[20] R Calabria andG Pulcini ldquoPoint estimation under asymmetricloss functions for left-truncated exponential samplesrdquo Commu-nications in StatisticsmdashTheory and Methods vol 25 no 3 pp585ndash600 1996
[21] A T K Wan G Zou and A H Lee ldquoMinimax and Γ-minimaxestimation for the Poisson distribution under LINEX loss whenthe parameter space is restrictedrdquo Statistics and ProbabilityLetters vol 50 no 1 pp 23ndash32 2000
[22] S Anatolyev ldquoKernel estimation under linear-exponential lossrdquoEconomics Letters vol 91 no 1 pp 39ndash43 2006
[23] X Li Y Shi J Wei and J Chai ldquoEmpirical Bayes estimators ofreliability performances using LINEX loss under progressivelytype-II censored samplesrdquo Mathematics and Computers inSimulation vol 73 no 5 pp 320ndash326 2007
[24] D K Dyer and P S L Liu ldquoOn comparison of estimators in ageneralized life modelrdquo Microelectronics Reliability vol 32 no1-2 pp 207ndash221 1992
[25] A A Soliman ldquoEstimation of parameters of life from progres-sively censored data using Burr-XII modelrdquo IEEE Transactionson Reliability vol 54 no 1 pp 34ndash42 2005
[26] A A Soliman and G R Elkahlout ldquoBayes estimation of thelogistic distribution based on progressively censored samplesrdquoJournal of Applied Statistical Science vol 14 no 3-4 pp 281ndash2932005
[27] C Caroni ldquoThe correct ldquoball bearingsrdquo datardquo Lifetime DataAnalysis vol 8 no 4 pp 395ndash399 2002
[28] M Z Raqab and M T Madi ldquoBayesian prediction of the totaltime on test using doubly censored Rayleigh datardquo Journal ofStatistical Computation and Simulation vol 72 no 10 pp 781ndash789 2002
[29] W C Lee ldquoStatistical testing for assessing lifetime performanceIndex of the Rayleigh lifetime productsrdquo Journal of the ChineseInstitute of Industrial Engineers vol 25 pp 433ndash445 2008
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
8 Journal of Applied Mathematics
where 119879 = 1198832
119880(119899)+ 119886 119886 and 119887 are the parameters of prior
distribution with density as (15)We construct a statistical testing procedure to assess
whether the lifetime performance index adheres to therequired level The one-sided credible confidence interval for119862119871is obtained using the pivotal quantity 1199051205792|
119909sim
=(119909119880(1)119909119880(119899))
By using the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899)) given the
specified significance level 120572 the level 100(1minus120572) one-sidedcredible interval for 119862
119871can be derived as follows
Since the pivotal quantity 1199051205792|119909sim
=(119909119880(1)119909119880(119899))sim 1205942
2(119899+119887)
and 12059422(119899+119887)1minus120572
function which represents the lower 1 minus 120572percentile of 1205942
2(119899+119887) then
119875(119905
1205792le 1205942
2(119899+119887)1minus120572| 119909sim) = 1 minus 120572
where 119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899))
997904rArr 119875(119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times[Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
| 119909sim)
= 1 minus 120572
(42)
where 119862119871as the definition of (5)
From (42) we obtain that a 100(1 minus 120572) one-sidedcredible interval for 119862
119871is
119862119871ge radic
120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times [Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
(43)
where 119862119871BG is given by (41)
Therefore the 100(1 minus 120572) lower credible interval boundfor 119862119871can be written as
LBBG = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times [Γ (119899 + 119887 + 1199022)
Γ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
(44)
where 119862119871BG is given by (41) 120572 is the specified significance
level and 119887 is a parameter of prior distribution with densityas (15)
The managers can use the one-sided credible interval todetermine whether the product performance attains to therequired level The proposed testing procedure of 119862
119871with
119862119871BG can be organized as follows
Step 1 Determine the lower lifetime limit 119871 for productsand the performance index value 119888
0 then the testing null
hypothesis 1198670 119862119871le 1198880and the alternative hypothesis
1198671 119862119871gt 1198880are constructed
Step 2 Specify a significance level 120572
Step 3 Given the parameters 119886 and 119887 of prior distributionthe parameter 119902 of GE loss function the upper record values119909sim= (119909119880(1) 119909119880(2) 119909
119880(119899)) the lower lifetime limit 119871 and the
significance level 120572 then we can calculate the 100(1 minus 120572)one-sided credible interval [LBBGinfin) for 119862119871 where LBBG asthe definition of (44)
Step 4 The decision rule of statistical test is provided asfollows
(1) if the performance index value 1198880notin [LBBGinfin) then
we will reject 1198670 It is concluded that the lifetime
performance index of products meets the requiredlevel
(2) if the performance index value 1198880isin [LBBGinfin) then
we do not reject 1198670 It is concluded that the lifetime
performance index of products does not meet therequired level
6 The Monte Carlo Simulation Algorithm ofConfidence (or Credible) Level
In this section we will report the results of a simulationstudy for confidence (or credible) level (1 minus 120572) based on a100(1 minus 120572) one-sided confidence (or credible) interval ofthe lifetime performance index 119862
119871 We considered 120572 = 005
and then generated samples from the Rayleigh distributionwith pdf as in (2) with respect to the record values
(i) The Monte Carlo simulation algorithm of confidencelevel (1 minus 120572) for 119862
119871under MLE is given in the
following steps
Step 1 Given 119899 119886 119887 119871 and 120572 where 119886 gt 0 119887 gt 0
Step 2 For the values of prior parameters (119886 119887) use (15) togenerate 120579 from the square-root inverted-gamma distribu-tion
Step 3(a) The generation of data (119885
1 1198852 119885
119899) is by standard
exponential distribution with parameter 1(b) Set1198842
1= 212057921198851and1198842119894= 21205792119885119894+1198842
119894minus1 for 119894 = 2 119899
1198841= radic21205792119885
1 119884
119894= radic21205792119885
119894+ 1198842
119894minus1 for 119894 = 2 119899
(45)
(1198841 1198842 119884
119899) are the upper record values from the
Rayleigh distribution(c) The value of LBMLE is calculated by
LBMLE = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871MLE)(
1205942
21198991minus120572
2119899)
12
(46)
Journal of Applied Mathematics 9
where 119862119871MLE is given by (11) and 1205942
21198991minus120572function
which represents the lower 1 minus 120572 percentile of 12059422119899
(d) If 119862119871ge LBMLE then count = 1 else count = 0
Step 4
(a) Step 3 is repeated 1000 times(b) The estimation of confidence level (1 minus 120572) is (1 minus 120572) =
(total count)1000
Step 5
(a) Repeat Step 2ndashStep 4 for 100 times then we can getthe 100 estimations of confidence level as follows
(1 minus 120572)1 (1 minus 120572)
2 (1 minus 120572)
100 (47)
(b) The average empirical confidence level 1 minus 120572
of (1 minus 120572)119894 119894 = 1 100 that is 1 minus 120572 =
(1100)sum100
119894=1(1 minus 120572)
119894
(c) The sample mean square error (SMSE) of(1 minus 120572)
1 (1 minus 120572)
2 (1 minus 120572)
100SMSE = (1100)
times sum100
119894=1[(1 minus 120572)
119894minus (1 minus 120572)]
2
The results of simulation are summarized in Table 2 basedon 119871 = 10 the different value of sample size 119899 priorparameter (119886 119887) = (2 5) (6 15) and (2 2) at 120572 = 005respectively The scope of SMSE is between 000413 and000593
Step 1 Given 119899 119886 119887 119871 and 120572 where 119886 gt 0 119887 gt 0
Step 2 For the values of prior parameters (119886 119887) use (15) togenerate 120579 from the square-root inverted-gamma distribu-tion
Step 3
(a) The generation of data (1198851 1198852 119885
119899) is by standard
exponential distribution with parameter 1(b) Set1198842
1= 212057921198851and1198842119894= 21205792119885119894+1198842
119894minus1 for 119894 = 2 119899
1198841= radic21205792119885
1 119884
119894= radic21205792119885
119894+ 1198842
119894minus1 for 119894 = 2 119899
(48)
(1198841 1198842 119884
119899) are the upper record values from the
Rayleigh distribution(c) The values of LBBS LBBL and LBBG are calculated by
(25) (36) and (44) respectively(d) If119862
119871ge LB then count = 1 else count = 0 where LB =
LBBS or LBBL or LBBG
Step 4
(a) Step 3 is repeated 1000 times(b) The estimation of credible level (1 minus 120572) is (1 minus 120572) =
(total count)1000
Step 5
(a) Repeat Step 2ndashStep 4 for 100 times then we can getthe 100 estimations of credible level as follows
(1 minus 120572)1 (1 minus 120572)
2 (1 minus 120572)
100 (49)
(b) The average empirical credible level 1 minus 120572 of (1 minus 120572)119894
119894 = 1 100 that is 1 minus 120572 = (1100)sum100119894=1(1 minus 120572)
119894
(c) The sample mean square error (SMSE) of(1 minus 120572)
1 (1 minus 120572)
2 (1 minus 120572)
100SMSE = (1100)
times sum100
119894=1[(1 minus 120572)
119894minus (1 minus 120572)]
2
Based on 119871 = 10 the different value of sample size119899 prior parameter (119886 119887) = (2 5) (6 15) (2 2) and 120572 =
005 the results of simulation are summarized in Tables 3ndash5under SE loss function LINEX loss function with parameter119888 = minus05 05 15 and GE loss function with parameter119902 = 3 5 8 respectively The following points can be drawn
(a) All of the SMSEs are small enough and the scope ofSMSE is between 000346 and 000467
(b) Fix the size 119899 comparison of prior parameter (119886 119887) =(2 5) (6 15) and (2 2) as follows
(i) the values of SMSEwith prior parameter (119886 119887) =(6 15) and (2 5) are smaller than (119886 119887) = (2 2)respectively
(ii) a comparison of prior parameter (119886 119887) = (2 5)and (2 2) that is fix prior parameter 119886 if priorparameter 119887 increases then it can be seen thatthe SMSE will decrease
(c) In Table 4 fix the size 119899 and the prior parameter (119886 119887)comparison the parameter of LINEX loss function 119888 =minus05 05 15 as followsthe values of SMSE with prior parameter 119888 = 05 15andminus15 are the same as the values of SMSE inTable 3
(d) In Table 5 fix the prior parameter (119886 119887) comparisonof GE loss function parameter 119902 = 3 5 8 as followsthe values of SMSE with prior parameter 119902 = 3 5 and8 are the same as the values of SMSE in Table 3
Hence these results from simulation studies illustrate thatthe performance of our proposed method is acceptableMoreover we suggest that the prior parameter (119886 119887) = (6 15)and (2 5) is appropriate for the square-root inverted-gammadistribution
7 Numerical Example
In this section we propose the new hypothesis testingprocedures to a real-life data In the following examplewe discuss a real-life data for 25 ball bearings to illustratethe use of the new hypothesis testing procedures in thelifetime performance of ball bearings The proposed testingprocedures not only can handle nonnormal lifetime data butalso can handle the upper record values
10 Journal of Applied Mathematics
Table 2 Average empirical confidence level (1 minus 120572) for 119862119871under MLE when 120572 = 005
119899 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
5 093653 (000462) 093482 (000467) 093577 (000507)10 093643 (000449) 093581 (000413) 093450 (000593)15 093523 (000489) 093557 (000503) 093463 (000516)119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 3 Average empirical credible level (1 minus 120572) for 119862119871under SE loss function when 120572 = 005
119899 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
5 094357 (000386) 093760 (000386) 093925 (000395)10 094244 (000359) 093704 (000359) 093664 (000462)15 094096 (000346) 093619 (000426) 093635 (000467)119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 4 Average empirical credible level (1 minus 120572) for 119862119871under LINEX loss function when 120572 = 005
119899 119888 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
minus05 094357 (000386) 093760 (000386) 093925 (000395)5 05 094357 (000386) 093760 (000386) 093925 (000395)
15 094357 (000386) 093760 (000386) 093925 (000395)minus05 094244 (000359) 093704 (000359) 093664 (000462)
10 05 094244 (000359) 093704 (000359) 093664 (000462)15 094244 (000359) 093704 (000359) 093664 (000462)minus05 094096 (000346) 093619 (000426) 093635 (000467)
15 05 094096 (000346) 093619 (000426) 093635 (000467)15 094096 (000346) 093619 (000426) 093635 (000467)
119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 5 Average empirical credible level (1 minus 120572) for 119862119871under GE loss function when 120572 = 005
119899 119902 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
3 094357 (000386) 093760 (000386) 093925 (000395)5 5 094357 (000386) 093760 (000386) 093925 (000395)
8 094357 (000386) 093760 (000386) 093925 (000395)3 094244 (000359) 093704 (000359) 093664 (000462)
10 5 094244 (000359) 093704 (000359) 093664 (000462)8 094244 (000359) 093704 (000359) 093664 (000462)3 094096 (000346) 093619 (000426) 093635 (000467)
15 5 094096 (000346) 093619 (000426) 093635 (000467)8 094096 (000346) 093619 (000426) 093635 (000467)
119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Example The data is the failure times of 25 ball bearings inendurance testThe 25 observations are the number ofmillionrevolutions before failure for each of the ball bearings Thedata come from Caroni [27] as follows
6780 6780 6780 6864 3300 6864 9864 12804 42122892 4560 5184 5556 17340 4848 1788 9312 54124152 5196 12792 8412 10512 10584 and 6888
Raqab and Madi [28] and Lee [29] indicated that a one-parameter Rayleigh distribution is acceptable for these dataIn addition we also have a way to test the hypothesis thatthe failure data come from the Rayleigh distribution Thetesting hypothesis119867
0 119883 comes from a Rayleigh distribution
versus1198671 119883 does not come from a Rayleigh distribution is
constructed under significance level 120572 = 005From a probability plot of Figure 1 by using the Minitab
Statistical Software we can conclude the operational lifetimesdata of 25 ball bearings from the Rayleigh distribution whichis the Weibull distribution with the shape 2 (also see Lee etal [14]) For informative prior we use the prior information119864(120579) = 120579MLE ≐ 54834 and Var(120579) = 02 giving the priorparameter values as (119886 = 6014 119887 = 1001)
Here if only the upper record values have been observedthese are 119909
119880(119894) 119894 = 1 5(= 119899) = 6780 6864 9864
12804 17340
Journal of Applied Mathematics 11
10010
99
90807060504030
20
10
5
32
1
Xi
Probability plot of XiWeibull-95 CI
()
Shape 2
Scale 8002N 25
AD 0431P -value 057
Figure 1 Probability plot for failure times of 25 ball bearings data
(1) The proposed testing procedure of 119862119871with 119862
119871MLE isstated as follows
Step 1 The record values from the above data are (119909119880(119894) 119894 =
1 2 3 4 5) = (6780 6864 9864 12804 17340)
Step 2 The lower specification limit 119871 is assumed to be 2337The deal with the product managersrsquo concerns regardinglifetime performance and the conforming rate 119875
119903of products
is required to exceed 80 percent Referring to Table 1 119862119871is
required to exceed 090Thus the performance index value isset at 119888
0= 090 The testing hypothesis119867
0 119862119871le 090 versus
1198671 119862119871gt 090 is constructed
Step 3 Specify a significance level 120572 = 005
Step 4 We can calculate that the 95 lower confidenceinterval bound for 119862
119871 where
LBMLE = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871MLE)(
1205942
21198991minus120572
2119899)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 126251254)(
1205942
10095
10)
12
= 103285
(50)So the 95 one-sided confidence interval for 119862
119871is
[LBMLEinfin) = [103285infin)
Step 5 Because of the performance index value 1198880= 090 notin
[LBMLEinfin) we reject the null hypothesis1198670 119862119871 le 090
Thus we can conclude that the lifetime performanceindex of data meets the required level
(2) The proposed testing procedures of 119862119871with 119862
119871BS119862119871BL and 119862119871BG are stated as follows
Step 1 The record values from the above data are (119909119880(119894) 119894 =
1 2 3 4 5) = (6780 6864 9864 12804 17340)
Step 2 The lower specification limit 119871 is assumed to be 2337The deal with the product managersrsquo concerns regardinglifetime performance and the conforming rate 119875
119903of products
is required to exceed 80 percent Referring to Table 1 119862119871is
required to exceed 090Thus the performance index value isset at 119888
0= 090 The testing hypothesis119867
0 119862119871le 090 versus
1198671 119862119871gt 090 is constructed
Step 3 Specify a significance level 120572 = 005 the parameter ofLINEX loss function 119888 = 05 and the parameter of GE lossfunction 119902 = 2
Step 4 We can calculate the 95 lower credible intervalbound for 119862
119871 where
LBBS = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BS)
timesΓ (119899 + 119887 minus 12)
Γ (119899 + 119887)(1205942
2(119899+119887)1minus120572
2)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1246074685)
timesΓ (5 + 1001 minus 12)
Γ (5 + 1001)(1205942
2(5+1001)095
2)
12
= 096984
(51)
12 Journal of Applied Mathematics
So the 95 one-sided credible interval for 119862119871
is[LBBSinfin) = [096984infin)
LBBL = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1129561059)
times [1
2 (05)(1 minus 119890
minus05(5+1001+1)) 1205942
2(5+1001)095]
12
= 096984
(52)
So the 95 one-sided credible interval for 119862119871
is[LBBLinfin) = [096984infin)
LBBG = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times [Γ (119899 + 119887 + 1199022)
Τ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1200432998)
times [Γ (5 + 1001 + 22)
Γ (5 + 1001)]
minus12
(1205942
2(5+1001)095
2)
12
= 096984
(53)
So the 95 one-sided credible interval for 119862119871
is[LBBGinfin) = [096984infin)
Step 5 Because of the performance index value 1198880= 090 notin
[LBBSinfin) = [LBBLinfin) = [LBBGinfin) we reject the nullhypothesis119867
0 119862119871le 090
Thus we can conclude that the lifetime performance index ofdata meets the required level In addition by using the priorparameter values (119886 119887) = (6 15) based on the simulationstudies we also obtain that 119888
0= 090 notin [LBBSinfin) =
[LBBLinfin) = [LBBGinfin) = 094033 So we also rejectthe null hypothesis 119867
0 119862119871le 090 Hence the lifetime
performance index of data meets the required level
8 Conclusions
Montgomery [1] proposed a process capability index 119862119871for
larger-the-better quality characteristic The assumption ofmost process capability is normal distribution but it is ofteninvalid In this paper we consider that Rayleigh distributionis the special case of Weibull distribution
Moreover in lifetime data testing experiments the exper-imenter may not always be in a position to observe the life
times of all the products put to test In industry and reliabilitystudies many products fail under stress for example anelectronic component ceases to function in an environmentof too high temperature and a battery dies under the stressof time So in such experiments measurement may be madesequentially and only the record values are observed
In order to let the process capability indices can beeffectively used This study constructs MLE and Bayesianestimator of 119862
119871under assuming the conjugate prior distribu-
tion and SE loss function LINEX loss function and GE lossfunction based on the upper record values from the Rayleighdistribution The Bayesian estimator of 119862
119871is then utilized to
develop a credible interval in the condition of known 119871This study also provides a table of the lifetime perfor-
mance index with its corresponding conforming rate basedon the Rayleigh distributionThat is the conforming rate andthe corresponding 119862
119871can be obtained
Further the Bayesian estimator and posterior distributionof 119862119871are also utilized to construct the testing procedure of
119862119871which is based on a credible interval If you want to test
whether the products meet the required level you can utilizethe proposed testing procedure which is easily applied andan effectively method to test in the condition of known 119871Numerical example is illustrated to show that the proposedtesting procedure is effectively evaluating whether the trueperformance index meets requirements
In addition these results from simulation studies illus-trate that the performance of our proposedmethod is accept-able According to SMSE the Bayesian approach is smallerthan the non-Bayesian approach we suggest that the Bayesianapproach is better than the non-Bayesian approachThen theSMSEs of prior parameters (119886 119887) = (6 15) and (2 5) aresmaller than the SMSE of prior parameter (119886 119887) = (2 2)We suggest that the prior parameters (119886 119887) = (6 15) and(2 5) are appropriate for the square-root inverted-gammadistribution based on the Bayesian estimators under theRayleigh distribution
Acknowledgments
The authors thank the referees for their suggestions andhelpful comments on revising the paper This research waspartially supported by the National Science Council China(Plans nos NSC 101-2221-E-158-003 NSC 99-2118-M-415-001 and NSC 100-2118-M-415-002) Moreover J W Wu andC W Hong have a direct financial relation with the NationalScience Council Taiwan but other authors do not a directfinancial relation with the National Science Council Taiwan
References
[1] D C Montgomery Introduction To Statistical Quality ControlJohn Wiley amp Sons New York NY USA 1985
[2] L I Tong K S Chen and H T Chen ldquoStatistical testingfor assessing the performance of lifetime index of electroniccomponents with exponential distributionrdquo Journal of QualityReliability Management vol 19 pp 812ndash824 2002
[3] C-W Hong J-W Wu and C-H Cheng ldquoComputationalprocedure of performance assessment of lifetime index of
Journal of Applied Mathematics 13
businesses for the Pareto lifetime model with the right type IIcensored samplerdquo Applied Mathematics and Computation vol184 no 2 pp 336ndash350 2007
[4] C W Hong J W Wu and C H Cheng ldquoComputational pro-cedure of performance assessment of lifetime index of Paretolifetime businesses based on confidence intervalrdquo Applied SoftComputing vol 8 no 1 pp 698ndash705 2008
[5] J-W Wu H-M Lee and C-L Lei ldquoComputational testingalgorithmic procedure of assessment for lifetime performanceindex of products with two-parameter exponential distribu-tionrdquo Applied Mathematics and Computation vol 190 no 1 pp116ndash125 2007
[6] W-C Lee J-W Wu and C-W Hong ldquoAssessing the lifetimeperformance index of products with the exponential distribu-tion under progressively type II right censored samplesrdquo Journalof Computational and Applied Mathematics vol 231 no 2 pp648ndash656 2009
[7] W C Lee J W Wu and C L Lei ldquoEvaluating the lifetimeperformance index for the exponential lifetime productsrdquoApplied Mathematical Modelling vol 34 no 5 pp 1217ndash12242010
[8] K N Chandler ldquoThe distribution and frequency of recordvaluesrdquo Journal of the Royal Statistical Society B vol 14 pp 220ndash228 1952
[9] S K Bhattacharya and R K Tyagi ldquoBayesian survival analysisbased on the rayleigh modelrdquo Trabajos de Estadistica vol 5 no1 pp 81ndash92 1990
[10] A D Cliff and J K Ord ldquoModel building and the analysis ofspatial pattern in human geography (with discussion)rdquo Journalof the Royal Statistical Society B vol 37 no 3 pp 297ndash348 1975
[11] D D Dyer and C W Whisenand ldquoBest linear unbiasedestimator of the parameter of the Rayleigh distribution I Smallsample theory for censored order statisticsrdquo IEEE Transactionson Reliability vol R-22 no 1 pp 27ndash34 1973
[12] A A Soliman and F M Al-Aboud ldquoBayesian inference usingrecord values from Rayleigh model with applicationrdquo EuropeanJournal of Operational Research vol 185 no 2 pp 659ndash6722008
[13] StatSoft STATISTICA Software Version 7 StatSoft Tulsa Okla2005
[14] W-C Lee J-W Wu M-L Hong L-S Lin and R-L ChanldquoAssessing the lifetime performance index of Rayleigh productsbased on the Bayesian estimation under progressive type IIright censored samplesrdquo Journal of Computational and AppliedMathematics vol 235 no 6 pp 1676ndash1688 2011
[15] M Ahsanullah Record Statistics Nova Science New York NYUSA 1995
[16] B C Arnold N Balakrishnan and H N Nagaraja RecordsJohn Wiley amp Sons New York NY USA 1998
[17] P W Zehna ldquoInvariance of maximum likelihood estimatorsrdquoAnnals of Mathematical Statistics vol 37 p 744 1966
[18] A Zellner ldquoBayesian estimation and prediction using asymmet-ric loss functionsrdquo Journal of the American Statistical Associa-tion vol 81 no 394 pp 446ndash451 1986
[19] S-Y Huang ldquoEmpirical Bayes testing procedures in somenonexponential families using asymmetric linex loss functionrdquoJournal of Statistical Planning and Inference vol 46 no 3 pp293ndash309 1995
[20] R Calabria andG Pulcini ldquoPoint estimation under asymmetricloss functions for left-truncated exponential samplesrdquo Commu-nications in StatisticsmdashTheory and Methods vol 25 no 3 pp585ndash600 1996
[21] A T K Wan G Zou and A H Lee ldquoMinimax and Γ-minimaxestimation for the Poisson distribution under LINEX loss whenthe parameter space is restrictedrdquo Statistics and ProbabilityLetters vol 50 no 1 pp 23ndash32 2000
[22] S Anatolyev ldquoKernel estimation under linear-exponential lossrdquoEconomics Letters vol 91 no 1 pp 39ndash43 2006
[23] X Li Y Shi J Wei and J Chai ldquoEmpirical Bayes estimators ofreliability performances using LINEX loss under progressivelytype-II censored samplesrdquo Mathematics and Computers inSimulation vol 73 no 5 pp 320ndash326 2007
[24] D K Dyer and P S L Liu ldquoOn comparison of estimators in ageneralized life modelrdquo Microelectronics Reliability vol 32 no1-2 pp 207ndash221 1992
[25] A A Soliman ldquoEstimation of parameters of life from progres-sively censored data using Burr-XII modelrdquo IEEE Transactionson Reliability vol 54 no 1 pp 34ndash42 2005
[26] A A Soliman and G R Elkahlout ldquoBayes estimation of thelogistic distribution based on progressively censored samplesrdquoJournal of Applied Statistical Science vol 14 no 3-4 pp 281ndash2932005
[27] C Caroni ldquoThe correct ldquoball bearingsrdquo datardquo Lifetime DataAnalysis vol 8 no 4 pp 395ndash399 2002
[28] M Z Raqab and M T Madi ldquoBayesian prediction of the totaltime on test using doubly censored Rayleigh datardquo Journal ofStatistical Computation and Simulation vol 72 no 10 pp 781ndash789 2002
[29] W C Lee ldquoStatistical testing for assessing lifetime performanceIndex of the Rayleigh lifetime productsrdquo Journal of the ChineseInstitute of Industrial Engineers vol 25 pp 433ndash445 2008
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Journal of Applied Mathematics 9
where 119862119871MLE is given by (11) and 1205942
21198991minus120572function
which represents the lower 1 minus 120572 percentile of 12059422119899
(d) If 119862119871ge LBMLE then count = 1 else count = 0
Step 4
(a) Step 3 is repeated 1000 times(b) The estimation of confidence level (1 minus 120572) is (1 minus 120572) =
(total count)1000
Step 5
(a) Repeat Step 2ndashStep 4 for 100 times then we can getthe 100 estimations of confidence level as follows
(1 minus 120572)1 (1 minus 120572)
2 (1 minus 120572)
100 (47)
(b) The average empirical confidence level 1 minus 120572
of (1 minus 120572)119894 119894 = 1 100 that is 1 minus 120572 =
(1100)sum100
119894=1(1 minus 120572)
119894
(c) The sample mean square error (SMSE) of(1 minus 120572)
1 (1 minus 120572)
2 (1 minus 120572)
100SMSE = (1100)
times sum100
119894=1[(1 minus 120572)
119894minus (1 minus 120572)]
2
The results of simulation are summarized in Table 2 basedon 119871 = 10 the different value of sample size 119899 priorparameter (119886 119887) = (2 5) (6 15) and (2 2) at 120572 = 005respectively The scope of SMSE is between 000413 and000593
Step 1 Given 119899 119886 119887 119871 and 120572 where 119886 gt 0 119887 gt 0
Step 2 For the values of prior parameters (119886 119887) use (15) togenerate 120579 from the square-root inverted-gamma distribu-tion
Step 3
(a) The generation of data (1198851 1198852 119885
119899) is by standard
exponential distribution with parameter 1(b) Set1198842
1= 212057921198851and1198842119894= 21205792119885119894+1198842
119894minus1 for 119894 = 2 119899
1198841= radic21205792119885
1 119884
119894= radic21205792119885
119894+ 1198842
119894minus1 for 119894 = 2 119899
(48)
(1198841 1198842 119884
119899) are the upper record values from the
Rayleigh distribution(c) The values of LBBS LBBL and LBBG are calculated by
(25) (36) and (44) respectively(d) If119862
119871ge LB then count = 1 else count = 0 where LB =
LBBS or LBBL or LBBG
Step 4
(a) Step 3 is repeated 1000 times(b) The estimation of credible level (1 minus 120572) is (1 minus 120572) =
(total count)1000
Step 5
(a) Repeat Step 2ndashStep 4 for 100 times then we can getthe 100 estimations of credible level as follows
(1 minus 120572)1 (1 minus 120572)
2 (1 minus 120572)
100 (49)
(b) The average empirical credible level 1 minus 120572 of (1 minus 120572)119894
119894 = 1 100 that is 1 minus 120572 = (1100)sum100119894=1(1 minus 120572)
119894
(c) The sample mean square error (SMSE) of(1 minus 120572)
1 (1 minus 120572)
2 (1 minus 120572)
100SMSE = (1100)
times sum100
119894=1[(1 minus 120572)
119894minus (1 minus 120572)]
2
Based on 119871 = 10 the different value of sample size119899 prior parameter (119886 119887) = (2 5) (6 15) (2 2) and 120572 =
005 the results of simulation are summarized in Tables 3ndash5under SE loss function LINEX loss function with parameter119888 = minus05 05 15 and GE loss function with parameter119902 = 3 5 8 respectively The following points can be drawn
(a) All of the SMSEs are small enough and the scope ofSMSE is between 000346 and 000467
(b) Fix the size 119899 comparison of prior parameter (119886 119887) =(2 5) (6 15) and (2 2) as follows
(i) the values of SMSEwith prior parameter (119886 119887) =(6 15) and (2 5) are smaller than (119886 119887) = (2 2)respectively
(ii) a comparison of prior parameter (119886 119887) = (2 5)and (2 2) that is fix prior parameter 119886 if priorparameter 119887 increases then it can be seen thatthe SMSE will decrease
(c) In Table 4 fix the size 119899 and the prior parameter (119886 119887)comparison the parameter of LINEX loss function 119888 =minus05 05 15 as followsthe values of SMSE with prior parameter 119888 = 05 15andminus15 are the same as the values of SMSE inTable 3
(d) In Table 5 fix the prior parameter (119886 119887) comparisonof GE loss function parameter 119902 = 3 5 8 as followsthe values of SMSE with prior parameter 119902 = 3 5 and8 are the same as the values of SMSE in Table 3
Hence these results from simulation studies illustrate thatthe performance of our proposed method is acceptableMoreover we suggest that the prior parameter (119886 119887) = (6 15)and (2 5) is appropriate for the square-root inverted-gammadistribution
7 Numerical Example
In this section we propose the new hypothesis testingprocedures to a real-life data In the following examplewe discuss a real-life data for 25 ball bearings to illustratethe use of the new hypothesis testing procedures in thelifetime performance of ball bearings The proposed testingprocedures not only can handle nonnormal lifetime data butalso can handle the upper record values
10 Journal of Applied Mathematics
Table 2 Average empirical confidence level (1 minus 120572) for 119862119871under MLE when 120572 = 005
119899 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
5 093653 (000462) 093482 (000467) 093577 (000507)10 093643 (000449) 093581 (000413) 093450 (000593)15 093523 (000489) 093557 (000503) 093463 (000516)119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 3 Average empirical credible level (1 minus 120572) for 119862119871under SE loss function when 120572 = 005
119899 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
5 094357 (000386) 093760 (000386) 093925 (000395)10 094244 (000359) 093704 (000359) 093664 (000462)15 094096 (000346) 093619 (000426) 093635 (000467)119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 4 Average empirical credible level (1 minus 120572) for 119862119871under LINEX loss function when 120572 = 005
119899 119888 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
minus05 094357 (000386) 093760 (000386) 093925 (000395)5 05 094357 (000386) 093760 (000386) 093925 (000395)
15 094357 (000386) 093760 (000386) 093925 (000395)minus05 094244 (000359) 093704 (000359) 093664 (000462)
10 05 094244 (000359) 093704 (000359) 093664 (000462)15 094244 (000359) 093704 (000359) 093664 (000462)minus05 094096 (000346) 093619 (000426) 093635 (000467)
15 05 094096 (000346) 093619 (000426) 093635 (000467)15 094096 (000346) 093619 (000426) 093635 (000467)
119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 5 Average empirical credible level (1 minus 120572) for 119862119871under GE loss function when 120572 = 005
119899 119902 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
3 094357 (000386) 093760 (000386) 093925 (000395)5 5 094357 (000386) 093760 (000386) 093925 (000395)
8 094357 (000386) 093760 (000386) 093925 (000395)3 094244 (000359) 093704 (000359) 093664 (000462)
10 5 094244 (000359) 093704 (000359) 093664 (000462)8 094244 (000359) 093704 (000359) 093664 (000462)3 094096 (000346) 093619 (000426) 093635 (000467)
15 5 094096 (000346) 093619 (000426) 093635 (000467)8 094096 (000346) 093619 (000426) 093635 (000467)
119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Example The data is the failure times of 25 ball bearings inendurance testThe 25 observations are the number ofmillionrevolutions before failure for each of the ball bearings Thedata come from Caroni [27] as follows
6780 6780 6780 6864 3300 6864 9864 12804 42122892 4560 5184 5556 17340 4848 1788 9312 54124152 5196 12792 8412 10512 10584 and 6888
Raqab and Madi [28] and Lee [29] indicated that a one-parameter Rayleigh distribution is acceptable for these dataIn addition we also have a way to test the hypothesis thatthe failure data come from the Rayleigh distribution Thetesting hypothesis119867
0 119883 comes from a Rayleigh distribution
versus1198671 119883 does not come from a Rayleigh distribution is
constructed under significance level 120572 = 005From a probability plot of Figure 1 by using the Minitab
Statistical Software we can conclude the operational lifetimesdata of 25 ball bearings from the Rayleigh distribution whichis the Weibull distribution with the shape 2 (also see Lee etal [14]) For informative prior we use the prior information119864(120579) = 120579MLE ≐ 54834 and Var(120579) = 02 giving the priorparameter values as (119886 = 6014 119887 = 1001)
Here if only the upper record values have been observedthese are 119909
119880(119894) 119894 = 1 5(= 119899) = 6780 6864 9864
12804 17340
Journal of Applied Mathematics 11
10010
99
90807060504030
20
10
5
32
1
Xi
Probability plot of XiWeibull-95 CI
()
Shape 2
Scale 8002N 25
AD 0431P -value 057
Figure 1 Probability plot for failure times of 25 ball bearings data
(1) The proposed testing procedure of 119862119871with 119862
119871MLE isstated as follows
Step 1 The record values from the above data are (119909119880(119894) 119894 =
1 2 3 4 5) = (6780 6864 9864 12804 17340)
Step 2 The lower specification limit 119871 is assumed to be 2337The deal with the product managersrsquo concerns regardinglifetime performance and the conforming rate 119875
119903of products
is required to exceed 80 percent Referring to Table 1 119862119871is
required to exceed 090Thus the performance index value isset at 119888
0= 090 The testing hypothesis119867
0 119862119871le 090 versus
1198671 119862119871gt 090 is constructed
Step 3 Specify a significance level 120572 = 005
Step 4 We can calculate that the 95 lower confidenceinterval bound for 119862
119871 where
LBMLE = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871MLE)(
1205942
21198991minus120572
2119899)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 126251254)(
1205942
10095
10)
12
= 103285
(50)So the 95 one-sided confidence interval for 119862
119871is
[LBMLEinfin) = [103285infin)
Step 5 Because of the performance index value 1198880= 090 notin
[LBMLEinfin) we reject the null hypothesis1198670 119862119871 le 090
Thus we can conclude that the lifetime performanceindex of data meets the required level
(2) The proposed testing procedures of 119862119871with 119862
119871BS119862119871BL and 119862119871BG are stated as follows
Step 1 The record values from the above data are (119909119880(119894) 119894 =
1 2 3 4 5) = (6780 6864 9864 12804 17340)
Step 2 The lower specification limit 119871 is assumed to be 2337The deal with the product managersrsquo concerns regardinglifetime performance and the conforming rate 119875
119903of products
is required to exceed 80 percent Referring to Table 1 119862119871is
required to exceed 090Thus the performance index value isset at 119888
0= 090 The testing hypothesis119867
0 119862119871le 090 versus
1198671 119862119871gt 090 is constructed
Step 3 Specify a significance level 120572 = 005 the parameter ofLINEX loss function 119888 = 05 and the parameter of GE lossfunction 119902 = 2
Step 4 We can calculate the 95 lower credible intervalbound for 119862
119871 where
LBBS = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BS)
timesΓ (119899 + 119887 minus 12)
Γ (119899 + 119887)(1205942
2(119899+119887)1minus120572
2)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1246074685)
timesΓ (5 + 1001 minus 12)
Γ (5 + 1001)(1205942
2(5+1001)095
2)
12
= 096984
(51)
12 Journal of Applied Mathematics
So the 95 one-sided credible interval for 119862119871
is[LBBSinfin) = [096984infin)
LBBL = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1129561059)
times [1
2 (05)(1 minus 119890
minus05(5+1001+1)) 1205942
2(5+1001)095]
12
= 096984
(52)
So the 95 one-sided credible interval for 119862119871
is[LBBLinfin) = [096984infin)
LBBG = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times [Γ (119899 + 119887 + 1199022)
Τ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1200432998)
times [Γ (5 + 1001 + 22)
Γ (5 + 1001)]
minus12
(1205942
2(5+1001)095
2)
12
= 096984
(53)
So the 95 one-sided credible interval for 119862119871
is[LBBGinfin) = [096984infin)
Step 5 Because of the performance index value 1198880= 090 notin
[LBBSinfin) = [LBBLinfin) = [LBBGinfin) we reject the nullhypothesis119867
0 119862119871le 090
Thus we can conclude that the lifetime performance index ofdata meets the required level In addition by using the priorparameter values (119886 119887) = (6 15) based on the simulationstudies we also obtain that 119888
0= 090 notin [LBBSinfin) =
[LBBLinfin) = [LBBGinfin) = 094033 So we also rejectthe null hypothesis 119867
0 119862119871le 090 Hence the lifetime
performance index of data meets the required level
8 Conclusions
Montgomery [1] proposed a process capability index 119862119871for
larger-the-better quality characteristic The assumption ofmost process capability is normal distribution but it is ofteninvalid In this paper we consider that Rayleigh distributionis the special case of Weibull distribution
Moreover in lifetime data testing experiments the exper-imenter may not always be in a position to observe the life
times of all the products put to test In industry and reliabilitystudies many products fail under stress for example anelectronic component ceases to function in an environmentof too high temperature and a battery dies under the stressof time So in such experiments measurement may be madesequentially and only the record values are observed
In order to let the process capability indices can beeffectively used This study constructs MLE and Bayesianestimator of 119862
119871under assuming the conjugate prior distribu-
tion and SE loss function LINEX loss function and GE lossfunction based on the upper record values from the Rayleighdistribution The Bayesian estimator of 119862
119871is then utilized to
develop a credible interval in the condition of known 119871This study also provides a table of the lifetime perfor-
mance index with its corresponding conforming rate basedon the Rayleigh distributionThat is the conforming rate andthe corresponding 119862
119871can be obtained
Further the Bayesian estimator and posterior distributionof 119862119871are also utilized to construct the testing procedure of
119862119871which is based on a credible interval If you want to test
whether the products meet the required level you can utilizethe proposed testing procedure which is easily applied andan effectively method to test in the condition of known 119871Numerical example is illustrated to show that the proposedtesting procedure is effectively evaluating whether the trueperformance index meets requirements
In addition these results from simulation studies illus-trate that the performance of our proposedmethod is accept-able According to SMSE the Bayesian approach is smallerthan the non-Bayesian approach we suggest that the Bayesianapproach is better than the non-Bayesian approachThen theSMSEs of prior parameters (119886 119887) = (6 15) and (2 5) aresmaller than the SMSE of prior parameter (119886 119887) = (2 2)We suggest that the prior parameters (119886 119887) = (6 15) and(2 5) are appropriate for the square-root inverted-gammadistribution based on the Bayesian estimators under theRayleigh distribution
Acknowledgments
The authors thank the referees for their suggestions andhelpful comments on revising the paper This research waspartially supported by the National Science Council China(Plans nos NSC 101-2221-E-158-003 NSC 99-2118-M-415-001 and NSC 100-2118-M-415-002) Moreover J W Wu andC W Hong have a direct financial relation with the NationalScience Council Taiwan but other authors do not a directfinancial relation with the National Science Council Taiwan
References
[1] D C Montgomery Introduction To Statistical Quality ControlJohn Wiley amp Sons New York NY USA 1985
[2] L I Tong K S Chen and H T Chen ldquoStatistical testingfor assessing the performance of lifetime index of electroniccomponents with exponential distributionrdquo Journal of QualityReliability Management vol 19 pp 812ndash824 2002
[3] C-W Hong J-W Wu and C-H Cheng ldquoComputationalprocedure of performance assessment of lifetime index of
Journal of Applied Mathematics 13
businesses for the Pareto lifetime model with the right type IIcensored samplerdquo Applied Mathematics and Computation vol184 no 2 pp 336ndash350 2007
[4] C W Hong J W Wu and C H Cheng ldquoComputational pro-cedure of performance assessment of lifetime index of Paretolifetime businesses based on confidence intervalrdquo Applied SoftComputing vol 8 no 1 pp 698ndash705 2008
[5] J-W Wu H-M Lee and C-L Lei ldquoComputational testingalgorithmic procedure of assessment for lifetime performanceindex of products with two-parameter exponential distribu-tionrdquo Applied Mathematics and Computation vol 190 no 1 pp116ndash125 2007
[6] W-C Lee J-W Wu and C-W Hong ldquoAssessing the lifetimeperformance index of products with the exponential distribu-tion under progressively type II right censored samplesrdquo Journalof Computational and Applied Mathematics vol 231 no 2 pp648ndash656 2009
[7] W C Lee J W Wu and C L Lei ldquoEvaluating the lifetimeperformance index for the exponential lifetime productsrdquoApplied Mathematical Modelling vol 34 no 5 pp 1217ndash12242010
[8] K N Chandler ldquoThe distribution and frequency of recordvaluesrdquo Journal of the Royal Statistical Society B vol 14 pp 220ndash228 1952
[9] S K Bhattacharya and R K Tyagi ldquoBayesian survival analysisbased on the rayleigh modelrdquo Trabajos de Estadistica vol 5 no1 pp 81ndash92 1990
[10] A D Cliff and J K Ord ldquoModel building and the analysis ofspatial pattern in human geography (with discussion)rdquo Journalof the Royal Statistical Society B vol 37 no 3 pp 297ndash348 1975
[11] D D Dyer and C W Whisenand ldquoBest linear unbiasedestimator of the parameter of the Rayleigh distribution I Smallsample theory for censored order statisticsrdquo IEEE Transactionson Reliability vol R-22 no 1 pp 27ndash34 1973
[12] A A Soliman and F M Al-Aboud ldquoBayesian inference usingrecord values from Rayleigh model with applicationrdquo EuropeanJournal of Operational Research vol 185 no 2 pp 659ndash6722008
[13] StatSoft STATISTICA Software Version 7 StatSoft Tulsa Okla2005
[14] W-C Lee J-W Wu M-L Hong L-S Lin and R-L ChanldquoAssessing the lifetime performance index of Rayleigh productsbased on the Bayesian estimation under progressive type IIright censored samplesrdquo Journal of Computational and AppliedMathematics vol 235 no 6 pp 1676ndash1688 2011
[15] M Ahsanullah Record Statistics Nova Science New York NYUSA 1995
[16] B C Arnold N Balakrishnan and H N Nagaraja RecordsJohn Wiley amp Sons New York NY USA 1998
[17] P W Zehna ldquoInvariance of maximum likelihood estimatorsrdquoAnnals of Mathematical Statistics vol 37 p 744 1966
[18] A Zellner ldquoBayesian estimation and prediction using asymmet-ric loss functionsrdquo Journal of the American Statistical Associa-tion vol 81 no 394 pp 446ndash451 1986
[19] S-Y Huang ldquoEmpirical Bayes testing procedures in somenonexponential families using asymmetric linex loss functionrdquoJournal of Statistical Planning and Inference vol 46 no 3 pp293ndash309 1995
[20] R Calabria andG Pulcini ldquoPoint estimation under asymmetricloss functions for left-truncated exponential samplesrdquo Commu-nications in StatisticsmdashTheory and Methods vol 25 no 3 pp585ndash600 1996
[21] A T K Wan G Zou and A H Lee ldquoMinimax and Γ-minimaxestimation for the Poisson distribution under LINEX loss whenthe parameter space is restrictedrdquo Statistics and ProbabilityLetters vol 50 no 1 pp 23ndash32 2000
[22] S Anatolyev ldquoKernel estimation under linear-exponential lossrdquoEconomics Letters vol 91 no 1 pp 39ndash43 2006
[23] X Li Y Shi J Wei and J Chai ldquoEmpirical Bayes estimators ofreliability performances using LINEX loss under progressivelytype-II censored samplesrdquo Mathematics and Computers inSimulation vol 73 no 5 pp 320ndash326 2007
[24] D K Dyer and P S L Liu ldquoOn comparison of estimators in ageneralized life modelrdquo Microelectronics Reliability vol 32 no1-2 pp 207ndash221 1992
[25] A A Soliman ldquoEstimation of parameters of life from progres-sively censored data using Burr-XII modelrdquo IEEE Transactionson Reliability vol 54 no 1 pp 34ndash42 2005
[26] A A Soliman and G R Elkahlout ldquoBayes estimation of thelogistic distribution based on progressively censored samplesrdquoJournal of Applied Statistical Science vol 14 no 3-4 pp 281ndash2932005
[27] C Caroni ldquoThe correct ldquoball bearingsrdquo datardquo Lifetime DataAnalysis vol 8 no 4 pp 395ndash399 2002
[28] M Z Raqab and M T Madi ldquoBayesian prediction of the totaltime on test using doubly censored Rayleigh datardquo Journal ofStatistical Computation and Simulation vol 72 no 10 pp 781ndash789 2002
[29] W C Lee ldquoStatistical testing for assessing lifetime performanceIndex of the Rayleigh lifetime productsrdquo Journal of the ChineseInstitute of Industrial Engineers vol 25 pp 433ndash445 2008
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
10 Journal of Applied Mathematics
Table 2 Average empirical confidence level (1 minus 120572) for 119862119871under MLE when 120572 = 005
119899 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
5 093653 (000462) 093482 (000467) 093577 (000507)10 093643 (000449) 093581 (000413) 093450 (000593)15 093523 (000489) 093557 (000503) 093463 (000516)119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 3 Average empirical credible level (1 minus 120572) for 119862119871under SE loss function when 120572 = 005
119899 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
5 094357 (000386) 093760 (000386) 093925 (000395)10 094244 (000359) 093704 (000359) 093664 (000462)15 094096 (000346) 093619 (000426) 093635 (000467)119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 4 Average empirical credible level (1 minus 120572) for 119862119871under LINEX loss function when 120572 = 005
119899 119888 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
minus05 094357 (000386) 093760 (000386) 093925 (000395)5 05 094357 (000386) 093760 (000386) 093925 (000395)
15 094357 (000386) 093760 (000386) 093925 (000395)minus05 094244 (000359) 093704 (000359) 093664 (000462)
10 05 094244 (000359) 093704 (000359) 093664 (000462)15 094244 (000359) 093704 (000359) 093664 (000462)minus05 094096 (000346) 093619 (000426) 093635 (000467)
15 05 094096 (000346) 093619 (000426) 093635 (000467)15 094096 (000346) 093619 (000426) 093635 (000467)
119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Table 5 Average empirical credible level (1 minus 120572) for 119862119871under GE loss function when 120572 = 005
119899 119902 119886 = 2 119887 = 5 119886 = 6 119887 = 15 119886 = 2 119887 = 2
3 094357 (000386) 093760 (000386) 093925 (000395)5 5 094357 (000386) 093760 (000386) 093925 (000395)
8 094357 (000386) 093760 (000386) 093925 (000395)3 094244 (000359) 093704 (000359) 093664 (000462)
10 5 094244 (000359) 093704 (000359) 093664 (000462)8 094244 (000359) 093704 (000359) 093664 (000462)3 094096 (000346) 093619 (000426) 093635 (000467)
15 5 094096 (000346) 093619 (000426) 093635 (000467)8 094096 (000346) 093619 (000426) 093635 (000467)
119899 denotes the sample size the values in parentheses are sample mean square error of (1 minus 120572)
Example The data is the failure times of 25 ball bearings inendurance testThe 25 observations are the number ofmillionrevolutions before failure for each of the ball bearings Thedata come from Caroni [27] as follows
6780 6780 6780 6864 3300 6864 9864 12804 42122892 4560 5184 5556 17340 4848 1788 9312 54124152 5196 12792 8412 10512 10584 and 6888
Raqab and Madi [28] and Lee [29] indicated that a one-parameter Rayleigh distribution is acceptable for these dataIn addition we also have a way to test the hypothesis thatthe failure data come from the Rayleigh distribution Thetesting hypothesis119867
0 119883 comes from a Rayleigh distribution
versus1198671 119883 does not come from a Rayleigh distribution is
constructed under significance level 120572 = 005From a probability plot of Figure 1 by using the Minitab
Statistical Software we can conclude the operational lifetimesdata of 25 ball bearings from the Rayleigh distribution whichis the Weibull distribution with the shape 2 (also see Lee etal [14]) For informative prior we use the prior information119864(120579) = 120579MLE ≐ 54834 and Var(120579) = 02 giving the priorparameter values as (119886 = 6014 119887 = 1001)
Here if only the upper record values have been observedthese are 119909
119880(119894) 119894 = 1 5(= 119899) = 6780 6864 9864
12804 17340
Journal of Applied Mathematics 11
10010
99
90807060504030
20
10
5
32
1
Xi
Probability plot of XiWeibull-95 CI
()
Shape 2
Scale 8002N 25
AD 0431P -value 057
Figure 1 Probability plot for failure times of 25 ball bearings data
(1) The proposed testing procedure of 119862119871with 119862
119871MLE isstated as follows
Step 1 The record values from the above data are (119909119880(119894) 119894 =
1 2 3 4 5) = (6780 6864 9864 12804 17340)
Step 2 The lower specification limit 119871 is assumed to be 2337The deal with the product managersrsquo concerns regardinglifetime performance and the conforming rate 119875
119903of products
is required to exceed 80 percent Referring to Table 1 119862119871is
required to exceed 090Thus the performance index value isset at 119888
0= 090 The testing hypothesis119867
0 119862119871le 090 versus
1198671 119862119871gt 090 is constructed
Step 3 Specify a significance level 120572 = 005
Step 4 We can calculate that the 95 lower confidenceinterval bound for 119862
119871 where
LBMLE = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871MLE)(
1205942
21198991minus120572
2119899)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 126251254)(
1205942
10095
10)
12
= 103285
(50)So the 95 one-sided confidence interval for 119862
119871is
[LBMLEinfin) = [103285infin)
Step 5 Because of the performance index value 1198880= 090 notin
[LBMLEinfin) we reject the null hypothesis1198670 119862119871 le 090
Thus we can conclude that the lifetime performanceindex of data meets the required level
(2) The proposed testing procedures of 119862119871with 119862
119871BS119862119871BL and 119862119871BG are stated as follows
Step 1 The record values from the above data are (119909119880(119894) 119894 =
1 2 3 4 5) = (6780 6864 9864 12804 17340)
Step 2 The lower specification limit 119871 is assumed to be 2337The deal with the product managersrsquo concerns regardinglifetime performance and the conforming rate 119875
119903of products
is required to exceed 80 percent Referring to Table 1 119862119871is
required to exceed 090Thus the performance index value isset at 119888
0= 090 The testing hypothesis119867
0 119862119871le 090 versus
1198671 119862119871gt 090 is constructed
Step 3 Specify a significance level 120572 = 005 the parameter ofLINEX loss function 119888 = 05 and the parameter of GE lossfunction 119902 = 2
Step 4 We can calculate the 95 lower credible intervalbound for 119862
119871 where
LBBS = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BS)
timesΓ (119899 + 119887 minus 12)
Γ (119899 + 119887)(1205942
2(119899+119887)1minus120572
2)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1246074685)
timesΓ (5 + 1001 minus 12)
Γ (5 + 1001)(1205942
2(5+1001)095
2)
12
= 096984
(51)
12 Journal of Applied Mathematics
So the 95 one-sided credible interval for 119862119871
is[LBBSinfin) = [096984infin)
LBBL = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1129561059)
times [1
2 (05)(1 minus 119890
minus05(5+1001+1)) 1205942
2(5+1001)095]
12
= 096984
(52)
So the 95 one-sided credible interval for 119862119871
is[LBBLinfin) = [096984infin)
LBBG = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times [Γ (119899 + 119887 + 1199022)
Τ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1200432998)
times [Γ (5 + 1001 + 22)
Γ (5 + 1001)]
minus12
(1205942
2(5+1001)095
2)
12
= 096984
(53)
So the 95 one-sided credible interval for 119862119871
is[LBBGinfin) = [096984infin)
Step 5 Because of the performance index value 1198880= 090 notin
[LBBSinfin) = [LBBLinfin) = [LBBGinfin) we reject the nullhypothesis119867
0 119862119871le 090
Thus we can conclude that the lifetime performance index ofdata meets the required level In addition by using the priorparameter values (119886 119887) = (6 15) based on the simulationstudies we also obtain that 119888
0= 090 notin [LBBSinfin) =
[LBBLinfin) = [LBBGinfin) = 094033 So we also rejectthe null hypothesis 119867
0 119862119871le 090 Hence the lifetime
performance index of data meets the required level
8 Conclusions
Montgomery [1] proposed a process capability index 119862119871for
larger-the-better quality characteristic The assumption ofmost process capability is normal distribution but it is ofteninvalid In this paper we consider that Rayleigh distributionis the special case of Weibull distribution
Moreover in lifetime data testing experiments the exper-imenter may not always be in a position to observe the life
times of all the products put to test In industry and reliabilitystudies many products fail under stress for example anelectronic component ceases to function in an environmentof too high temperature and a battery dies under the stressof time So in such experiments measurement may be madesequentially and only the record values are observed
In order to let the process capability indices can beeffectively used This study constructs MLE and Bayesianestimator of 119862
119871under assuming the conjugate prior distribu-
tion and SE loss function LINEX loss function and GE lossfunction based on the upper record values from the Rayleighdistribution The Bayesian estimator of 119862
119871is then utilized to
develop a credible interval in the condition of known 119871This study also provides a table of the lifetime perfor-
mance index with its corresponding conforming rate basedon the Rayleigh distributionThat is the conforming rate andthe corresponding 119862
119871can be obtained
Further the Bayesian estimator and posterior distributionof 119862119871are also utilized to construct the testing procedure of
119862119871which is based on a credible interval If you want to test
whether the products meet the required level you can utilizethe proposed testing procedure which is easily applied andan effectively method to test in the condition of known 119871Numerical example is illustrated to show that the proposedtesting procedure is effectively evaluating whether the trueperformance index meets requirements
In addition these results from simulation studies illus-trate that the performance of our proposedmethod is accept-able According to SMSE the Bayesian approach is smallerthan the non-Bayesian approach we suggest that the Bayesianapproach is better than the non-Bayesian approachThen theSMSEs of prior parameters (119886 119887) = (6 15) and (2 5) aresmaller than the SMSE of prior parameter (119886 119887) = (2 2)We suggest that the prior parameters (119886 119887) = (6 15) and(2 5) are appropriate for the square-root inverted-gammadistribution based on the Bayesian estimators under theRayleigh distribution
Acknowledgments
The authors thank the referees for their suggestions andhelpful comments on revising the paper This research waspartially supported by the National Science Council China(Plans nos NSC 101-2221-E-158-003 NSC 99-2118-M-415-001 and NSC 100-2118-M-415-002) Moreover J W Wu andC W Hong have a direct financial relation with the NationalScience Council Taiwan but other authors do not a directfinancial relation with the National Science Council Taiwan
References
[1] D C Montgomery Introduction To Statistical Quality ControlJohn Wiley amp Sons New York NY USA 1985
[2] L I Tong K S Chen and H T Chen ldquoStatistical testingfor assessing the performance of lifetime index of electroniccomponents with exponential distributionrdquo Journal of QualityReliability Management vol 19 pp 812ndash824 2002
[3] C-W Hong J-W Wu and C-H Cheng ldquoComputationalprocedure of performance assessment of lifetime index of
Journal of Applied Mathematics 13
businesses for the Pareto lifetime model with the right type IIcensored samplerdquo Applied Mathematics and Computation vol184 no 2 pp 336ndash350 2007
[4] C W Hong J W Wu and C H Cheng ldquoComputational pro-cedure of performance assessment of lifetime index of Paretolifetime businesses based on confidence intervalrdquo Applied SoftComputing vol 8 no 1 pp 698ndash705 2008
[5] J-W Wu H-M Lee and C-L Lei ldquoComputational testingalgorithmic procedure of assessment for lifetime performanceindex of products with two-parameter exponential distribu-tionrdquo Applied Mathematics and Computation vol 190 no 1 pp116ndash125 2007
[6] W-C Lee J-W Wu and C-W Hong ldquoAssessing the lifetimeperformance index of products with the exponential distribu-tion under progressively type II right censored samplesrdquo Journalof Computational and Applied Mathematics vol 231 no 2 pp648ndash656 2009
[7] W C Lee J W Wu and C L Lei ldquoEvaluating the lifetimeperformance index for the exponential lifetime productsrdquoApplied Mathematical Modelling vol 34 no 5 pp 1217ndash12242010
[8] K N Chandler ldquoThe distribution and frequency of recordvaluesrdquo Journal of the Royal Statistical Society B vol 14 pp 220ndash228 1952
[9] S K Bhattacharya and R K Tyagi ldquoBayesian survival analysisbased on the rayleigh modelrdquo Trabajos de Estadistica vol 5 no1 pp 81ndash92 1990
[10] A D Cliff and J K Ord ldquoModel building and the analysis ofspatial pattern in human geography (with discussion)rdquo Journalof the Royal Statistical Society B vol 37 no 3 pp 297ndash348 1975
[11] D D Dyer and C W Whisenand ldquoBest linear unbiasedestimator of the parameter of the Rayleigh distribution I Smallsample theory for censored order statisticsrdquo IEEE Transactionson Reliability vol R-22 no 1 pp 27ndash34 1973
[12] A A Soliman and F M Al-Aboud ldquoBayesian inference usingrecord values from Rayleigh model with applicationrdquo EuropeanJournal of Operational Research vol 185 no 2 pp 659ndash6722008
[13] StatSoft STATISTICA Software Version 7 StatSoft Tulsa Okla2005
[14] W-C Lee J-W Wu M-L Hong L-S Lin and R-L ChanldquoAssessing the lifetime performance index of Rayleigh productsbased on the Bayesian estimation under progressive type IIright censored samplesrdquo Journal of Computational and AppliedMathematics vol 235 no 6 pp 1676ndash1688 2011
[15] M Ahsanullah Record Statistics Nova Science New York NYUSA 1995
[16] B C Arnold N Balakrishnan and H N Nagaraja RecordsJohn Wiley amp Sons New York NY USA 1998
[17] P W Zehna ldquoInvariance of maximum likelihood estimatorsrdquoAnnals of Mathematical Statistics vol 37 p 744 1966
[18] A Zellner ldquoBayesian estimation and prediction using asymmet-ric loss functionsrdquo Journal of the American Statistical Associa-tion vol 81 no 394 pp 446ndash451 1986
[19] S-Y Huang ldquoEmpirical Bayes testing procedures in somenonexponential families using asymmetric linex loss functionrdquoJournal of Statistical Planning and Inference vol 46 no 3 pp293ndash309 1995
[20] R Calabria andG Pulcini ldquoPoint estimation under asymmetricloss functions for left-truncated exponential samplesrdquo Commu-nications in StatisticsmdashTheory and Methods vol 25 no 3 pp585ndash600 1996
[21] A T K Wan G Zou and A H Lee ldquoMinimax and Γ-minimaxestimation for the Poisson distribution under LINEX loss whenthe parameter space is restrictedrdquo Statistics and ProbabilityLetters vol 50 no 1 pp 23ndash32 2000
[22] S Anatolyev ldquoKernel estimation under linear-exponential lossrdquoEconomics Letters vol 91 no 1 pp 39ndash43 2006
[23] X Li Y Shi J Wei and J Chai ldquoEmpirical Bayes estimators ofreliability performances using LINEX loss under progressivelytype-II censored samplesrdquo Mathematics and Computers inSimulation vol 73 no 5 pp 320ndash326 2007
[24] D K Dyer and P S L Liu ldquoOn comparison of estimators in ageneralized life modelrdquo Microelectronics Reliability vol 32 no1-2 pp 207ndash221 1992
[25] A A Soliman ldquoEstimation of parameters of life from progres-sively censored data using Burr-XII modelrdquo IEEE Transactionson Reliability vol 54 no 1 pp 34ndash42 2005
[26] A A Soliman and G R Elkahlout ldquoBayes estimation of thelogistic distribution based on progressively censored samplesrdquoJournal of Applied Statistical Science vol 14 no 3-4 pp 281ndash2932005
[27] C Caroni ldquoThe correct ldquoball bearingsrdquo datardquo Lifetime DataAnalysis vol 8 no 4 pp 395ndash399 2002
[28] M Z Raqab and M T Madi ldquoBayesian prediction of the totaltime on test using doubly censored Rayleigh datardquo Journal ofStatistical Computation and Simulation vol 72 no 10 pp 781ndash789 2002
[29] W C Lee ldquoStatistical testing for assessing lifetime performanceIndex of the Rayleigh lifetime productsrdquo Journal of the ChineseInstitute of Industrial Engineers vol 25 pp 433ndash445 2008
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Journal of Applied Mathematics 11
10010
99
90807060504030
20
10
5
32
1
Xi
Probability plot of XiWeibull-95 CI
()
Shape 2
Scale 8002N 25
AD 0431P -value 057
Figure 1 Probability plot for failure times of 25 ball bearings data
(1) The proposed testing procedure of 119862119871with 119862
119871MLE isstated as follows
Step 1 The record values from the above data are (119909119880(119894) 119894 =
1 2 3 4 5) = (6780 6864 9864 12804 17340)
Step 2 The lower specification limit 119871 is assumed to be 2337The deal with the product managersrsquo concerns regardinglifetime performance and the conforming rate 119875
119903of products
is required to exceed 80 percent Referring to Table 1 119862119871is
required to exceed 090Thus the performance index value isset at 119888
0= 090 The testing hypothesis119867
0 119862119871le 090 versus
1198671 119862119871gt 090 is constructed
Step 3 Specify a significance level 120572 = 005
Step 4 We can calculate that the 95 lower confidenceinterval bound for 119862
119871 where
LBMLE = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871MLE)(
1205942
21198991minus120572
2119899)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 126251254)(
1205942
10095
10)
12
= 103285
(50)So the 95 one-sided confidence interval for 119862
119871is
[LBMLEinfin) = [103285infin)
Step 5 Because of the performance index value 1198880= 090 notin
[LBMLEinfin) we reject the null hypothesis1198670 119862119871 le 090
Thus we can conclude that the lifetime performanceindex of data meets the required level
(2) The proposed testing procedures of 119862119871with 119862
119871BS119862119871BL and 119862119871BG are stated as follows
Step 1 The record values from the above data are (119909119880(119894) 119894 =
1 2 3 4 5) = (6780 6864 9864 12804 17340)
Step 2 The lower specification limit 119871 is assumed to be 2337The deal with the product managersrsquo concerns regardinglifetime performance and the conforming rate 119875
119903of products
is required to exceed 80 percent Referring to Table 1 119862119871is
required to exceed 090Thus the performance index value isset at 119888
0= 090 The testing hypothesis119867
0 119862119871le 090 versus
1198671 119862119871gt 090 is constructed
Step 3 Specify a significance level 120572 = 005 the parameter ofLINEX loss function 119888 = 05 and the parameter of GE lossfunction 119902 = 2
Step 4 We can calculate the 95 lower credible intervalbound for 119862
119871 where
LBBS = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BS)
timesΓ (119899 + 119887 minus 12)
Γ (119899 + 119887)(1205942
2(119899+119887)1minus120572
2)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1246074685)
timesΓ (5 + 1001 minus 12)
Γ (5 + 1001)(1205942
2(5+1001)095
2)
12
= 096984
(51)
12 Journal of Applied Mathematics
So the 95 one-sided credible interval for 119862119871
is[LBBSinfin) = [096984infin)
LBBL = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1129561059)
times [1
2 (05)(1 minus 119890
minus05(5+1001+1)) 1205942
2(5+1001)095]
12
= 096984
(52)
So the 95 one-sided credible interval for 119862119871
is[LBBLinfin) = [096984infin)
LBBG = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times [Γ (119899 + 119887 + 1199022)
Τ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1200432998)
times [Γ (5 + 1001 + 22)
Γ (5 + 1001)]
minus12
(1205942
2(5+1001)095
2)
12
= 096984
(53)
So the 95 one-sided credible interval for 119862119871
is[LBBGinfin) = [096984infin)
Step 5 Because of the performance index value 1198880= 090 notin
[LBBSinfin) = [LBBLinfin) = [LBBGinfin) we reject the nullhypothesis119867
0 119862119871le 090
Thus we can conclude that the lifetime performance index ofdata meets the required level In addition by using the priorparameter values (119886 119887) = (6 15) based on the simulationstudies we also obtain that 119888
0= 090 notin [LBBSinfin) =
[LBBLinfin) = [LBBGinfin) = 094033 So we also rejectthe null hypothesis 119867
0 119862119871le 090 Hence the lifetime
performance index of data meets the required level
8 Conclusions
Montgomery [1] proposed a process capability index 119862119871for
larger-the-better quality characteristic The assumption ofmost process capability is normal distribution but it is ofteninvalid In this paper we consider that Rayleigh distributionis the special case of Weibull distribution
Moreover in lifetime data testing experiments the exper-imenter may not always be in a position to observe the life
times of all the products put to test In industry and reliabilitystudies many products fail under stress for example anelectronic component ceases to function in an environmentof too high temperature and a battery dies under the stressof time So in such experiments measurement may be madesequentially and only the record values are observed
In order to let the process capability indices can beeffectively used This study constructs MLE and Bayesianestimator of 119862
119871under assuming the conjugate prior distribu-
tion and SE loss function LINEX loss function and GE lossfunction based on the upper record values from the Rayleighdistribution The Bayesian estimator of 119862
119871is then utilized to
develop a credible interval in the condition of known 119871This study also provides a table of the lifetime perfor-
mance index with its corresponding conforming rate basedon the Rayleigh distributionThat is the conforming rate andthe corresponding 119862
119871can be obtained
Further the Bayesian estimator and posterior distributionof 119862119871are also utilized to construct the testing procedure of
119862119871which is based on a credible interval If you want to test
whether the products meet the required level you can utilizethe proposed testing procedure which is easily applied andan effectively method to test in the condition of known 119871Numerical example is illustrated to show that the proposedtesting procedure is effectively evaluating whether the trueperformance index meets requirements
In addition these results from simulation studies illus-trate that the performance of our proposedmethod is accept-able According to SMSE the Bayesian approach is smallerthan the non-Bayesian approach we suggest that the Bayesianapproach is better than the non-Bayesian approachThen theSMSEs of prior parameters (119886 119887) = (6 15) and (2 5) aresmaller than the SMSE of prior parameter (119886 119887) = (2 2)We suggest that the prior parameters (119886 119887) = (6 15) and(2 5) are appropriate for the square-root inverted-gammadistribution based on the Bayesian estimators under theRayleigh distribution
Acknowledgments
The authors thank the referees for their suggestions andhelpful comments on revising the paper This research waspartially supported by the National Science Council China(Plans nos NSC 101-2221-E-158-003 NSC 99-2118-M-415-001 and NSC 100-2118-M-415-002) Moreover J W Wu andC W Hong have a direct financial relation with the NationalScience Council Taiwan but other authors do not a directfinancial relation with the National Science Council Taiwan
References
[1] D C Montgomery Introduction To Statistical Quality ControlJohn Wiley amp Sons New York NY USA 1985
[2] L I Tong K S Chen and H T Chen ldquoStatistical testingfor assessing the performance of lifetime index of electroniccomponents with exponential distributionrdquo Journal of QualityReliability Management vol 19 pp 812ndash824 2002
[3] C-W Hong J-W Wu and C-H Cheng ldquoComputationalprocedure of performance assessment of lifetime index of
Journal of Applied Mathematics 13
businesses for the Pareto lifetime model with the right type IIcensored samplerdquo Applied Mathematics and Computation vol184 no 2 pp 336ndash350 2007
[4] C W Hong J W Wu and C H Cheng ldquoComputational pro-cedure of performance assessment of lifetime index of Paretolifetime businesses based on confidence intervalrdquo Applied SoftComputing vol 8 no 1 pp 698ndash705 2008
[5] J-W Wu H-M Lee and C-L Lei ldquoComputational testingalgorithmic procedure of assessment for lifetime performanceindex of products with two-parameter exponential distribu-tionrdquo Applied Mathematics and Computation vol 190 no 1 pp116ndash125 2007
[6] W-C Lee J-W Wu and C-W Hong ldquoAssessing the lifetimeperformance index of products with the exponential distribu-tion under progressively type II right censored samplesrdquo Journalof Computational and Applied Mathematics vol 231 no 2 pp648ndash656 2009
[7] W C Lee J W Wu and C L Lei ldquoEvaluating the lifetimeperformance index for the exponential lifetime productsrdquoApplied Mathematical Modelling vol 34 no 5 pp 1217ndash12242010
[8] K N Chandler ldquoThe distribution and frequency of recordvaluesrdquo Journal of the Royal Statistical Society B vol 14 pp 220ndash228 1952
[9] S K Bhattacharya and R K Tyagi ldquoBayesian survival analysisbased on the rayleigh modelrdquo Trabajos de Estadistica vol 5 no1 pp 81ndash92 1990
[10] A D Cliff and J K Ord ldquoModel building and the analysis ofspatial pattern in human geography (with discussion)rdquo Journalof the Royal Statistical Society B vol 37 no 3 pp 297ndash348 1975
[11] D D Dyer and C W Whisenand ldquoBest linear unbiasedestimator of the parameter of the Rayleigh distribution I Smallsample theory for censored order statisticsrdquo IEEE Transactionson Reliability vol R-22 no 1 pp 27ndash34 1973
[12] A A Soliman and F M Al-Aboud ldquoBayesian inference usingrecord values from Rayleigh model with applicationrdquo EuropeanJournal of Operational Research vol 185 no 2 pp 659ndash6722008
[13] StatSoft STATISTICA Software Version 7 StatSoft Tulsa Okla2005
[14] W-C Lee J-W Wu M-L Hong L-S Lin and R-L ChanldquoAssessing the lifetime performance index of Rayleigh productsbased on the Bayesian estimation under progressive type IIright censored samplesrdquo Journal of Computational and AppliedMathematics vol 235 no 6 pp 1676ndash1688 2011
[15] M Ahsanullah Record Statistics Nova Science New York NYUSA 1995
[16] B C Arnold N Balakrishnan and H N Nagaraja RecordsJohn Wiley amp Sons New York NY USA 1998
[17] P W Zehna ldquoInvariance of maximum likelihood estimatorsrdquoAnnals of Mathematical Statistics vol 37 p 744 1966
[18] A Zellner ldquoBayesian estimation and prediction using asymmet-ric loss functionsrdquo Journal of the American Statistical Associa-tion vol 81 no 394 pp 446ndash451 1986
[19] S-Y Huang ldquoEmpirical Bayes testing procedures in somenonexponential families using asymmetric linex loss functionrdquoJournal of Statistical Planning and Inference vol 46 no 3 pp293ndash309 1995
[20] R Calabria andG Pulcini ldquoPoint estimation under asymmetricloss functions for left-truncated exponential samplesrdquo Commu-nications in StatisticsmdashTheory and Methods vol 25 no 3 pp585ndash600 1996
[21] A T K Wan G Zou and A H Lee ldquoMinimax and Γ-minimaxestimation for the Poisson distribution under LINEX loss whenthe parameter space is restrictedrdquo Statistics and ProbabilityLetters vol 50 no 1 pp 23ndash32 2000
[22] S Anatolyev ldquoKernel estimation under linear-exponential lossrdquoEconomics Letters vol 91 no 1 pp 39ndash43 2006
[23] X Li Y Shi J Wei and J Chai ldquoEmpirical Bayes estimators ofreliability performances using LINEX loss under progressivelytype-II censored samplesrdquo Mathematics and Computers inSimulation vol 73 no 5 pp 320ndash326 2007
[24] D K Dyer and P S L Liu ldquoOn comparison of estimators in ageneralized life modelrdquo Microelectronics Reliability vol 32 no1-2 pp 207ndash221 1992
[25] A A Soliman ldquoEstimation of parameters of life from progres-sively censored data using Burr-XII modelrdquo IEEE Transactionson Reliability vol 54 no 1 pp 34ndash42 2005
[26] A A Soliman and G R Elkahlout ldquoBayes estimation of thelogistic distribution based on progressively censored samplesrdquoJournal of Applied Statistical Science vol 14 no 3-4 pp 281ndash2932005
[27] C Caroni ldquoThe correct ldquoball bearingsrdquo datardquo Lifetime DataAnalysis vol 8 no 4 pp 395ndash399 2002
[28] M Z Raqab and M T Madi ldquoBayesian prediction of the totaltime on test using doubly censored Rayleigh datardquo Journal ofStatistical Computation and Simulation vol 72 no 10 pp 781ndash789 2002
[29] W C Lee ldquoStatistical testing for assessing lifetime performanceIndex of the Rayleigh lifetime productsrdquo Journal of the ChineseInstitute of Industrial Engineers vol 25 pp 433ndash445 2008
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
12 Journal of Applied Mathematics
So the 95 one-sided credible interval for 119862119871
is[LBBSinfin) = [096984infin)
LBBL = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BL)
times [1
2119888(1 minus 119890
minus119888(119899+119887+1)) 1205942
2(119899+119887)1minus120572]
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1129561059)
times [1
2 (05)(1 minus 119890
minus05(5+1001+1)) 1205942
2(5+1001)095]
12
= 096984
(52)
So the 95 one-sided credible interval for 119862119871
is[LBBLinfin) = [096984infin)
LBBG = radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 119862119871BG)
times [Γ (119899 + 119887 + 1199022)
Τ (119899 + 119887)]
minus1119902
(1205942
2(119899+119887)1minus120572
2)
12
= radic120587
4 minus 120587minus (radic
120587
4 minus 120587minus 1200432998)
times [Γ (5 + 1001 + 22)
Γ (5 + 1001)]
minus12
(1205942
2(5+1001)095
2)
12
= 096984
(53)
So the 95 one-sided credible interval for 119862119871
is[LBBGinfin) = [096984infin)
Step 5 Because of the performance index value 1198880= 090 notin
[LBBSinfin) = [LBBLinfin) = [LBBGinfin) we reject the nullhypothesis119867
0 119862119871le 090
Thus we can conclude that the lifetime performance index ofdata meets the required level In addition by using the priorparameter values (119886 119887) = (6 15) based on the simulationstudies we also obtain that 119888
0= 090 notin [LBBSinfin) =
[LBBLinfin) = [LBBGinfin) = 094033 So we also rejectthe null hypothesis 119867
0 119862119871le 090 Hence the lifetime
performance index of data meets the required level
8 Conclusions
Montgomery [1] proposed a process capability index 119862119871for
larger-the-better quality characteristic The assumption ofmost process capability is normal distribution but it is ofteninvalid In this paper we consider that Rayleigh distributionis the special case of Weibull distribution
Moreover in lifetime data testing experiments the exper-imenter may not always be in a position to observe the life
times of all the products put to test In industry and reliabilitystudies many products fail under stress for example anelectronic component ceases to function in an environmentof too high temperature and a battery dies under the stressof time So in such experiments measurement may be madesequentially and only the record values are observed
In order to let the process capability indices can beeffectively used This study constructs MLE and Bayesianestimator of 119862
119871under assuming the conjugate prior distribu-
tion and SE loss function LINEX loss function and GE lossfunction based on the upper record values from the Rayleighdistribution The Bayesian estimator of 119862
119871is then utilized to
develop a credible interval in the condition of known 119871This study also provides a table of the lifetime perfor-
mance index with its corresponding conforming rate basedon the Rayleigh distributionThat is the conforming rate andthe corresponding 119862
119871can be obtained
Further the Bayesian estimator and posterior distributionof 119862119871are also utilized to construct the testing procedure of
119862119871which is based on a credible interval If you want to test
whether the products meet the required level you can utilizethe proposed testing procedure which is easily applied andan effectively method to test in the condition of known 119871Numerical example is illustrated to show that the proposedtesting procedure is effectively evaluating whether the trueperformance index meets requirements
In addition these results from simulation studies illus-trate that the performance of our proposedmethod is accept-able According to SMSE the Bayesian approach is smallerthan the non-Bayesian approach we suggest that the Bayesianapproach is better than the non-Bayesian approachThen theSMSEs of prior parameters (119886 119887) = (6 15) and (2 5) aresmaller than the SMSE of prior parameter (119886 119887) = (2 2)We suggest that the prior parameters (119886 119887) = (6 15) and(2 5) are appropriate for the square-root inverted-gammadistribution based on the Bayesian estimators under theRayleigh distribution
Acknowledgments
The authors thank the referees for their suggestions andhelpful comments on revising the paper This research waspartially supported by the National Science Council China(Plans nos NSC 101-2221-E-158-003 NSC 99-2118-M-415-001 and NSC 100-2118-M-415-002) Moreover J W Wu andC W Hong have a direct financial relation with the NationalScience Council Taiwan but other authors do not a directfinancial relation with the National Science Council Taiwan
References
[1] D C Montgomery Introduction To Statistical Quality ControlJohn Wiley amp Sons New York NY USA 1985
[2] L I Tong K S Chen and H T Chen ldquoStatistical testingfor assessing the performance of lifetime index of electroniccomponents with exponential distributionrdquo Journal of QualityReliability Management vol 19 pp 812ndash824 2002
[3] C-W Hong J-W Wu and C-H Cheng ldquoComputationalprocedure of performance assessment of lifetime index of
Journal of Applied Mathematics 13
businesses for the Pareto lifetime model with the right type IIcensored samplerdquo Applied Mathematics and Computation vol184 no 2 pp 336ndash350 2007
[4] C W Hong J W Wu and C H Cheng ldquoComputational pro-cedure of performance assessment of lifetime index of Paretolifetime businesses based on confidence intervalrdquo Applied SoftComputing vol 8 no 1 pp 698ndash705 2008
[5] J-W Wu H-M Lee and C-L Lei ldquoComputational testingalgorithmic procedure of assessment for lifetime performanceindex of products with two-parameter exponential distribu-tionrdquo Applied Mathematics and Computation vol 190 no 1 pp116ndash125 2007
[6] W-C Lee J-W Wu and C-W Hong ldquoAssessing the lifetimeperformance index of products with the exponential distribu-tion under progressively type II right censored samplesrdquo Journalof Computational and Applied Mathematics vol 231 no 2 pp648ndash656 2009
[7] W C Lee J W Wu and C L Lei ldquoEvaluating the lifetimeperformance index for the exponential lifetime productsrdquoApplied Mathematical Modelling vol 34 no 5 pp 1217ndash12242010
[8] K N Chandler ldquoThe distribution and frequency of recordvaluesrdquo Journal of the Royal Statistical Society B vol 14 pp 220ndash228 1952
[9] S K Bhattacharya and R K Tyagi ldquoBayesian survival analysisbased on the rayleigh modelrdquo Trabajos de Estadistica vol 5 no1 pp 81ndash92 1990
[10] A D Cliff and J K Ord ldquoModel building and the analysis ofspatial pattern in human geography (with discussion)rdquo Journalof the Royal Statistical Society B vol 37 no 3 pp 297ndash348 1975
[11] D D Dyer and C W Whisenand ldquoBest linear unbiasedestimator of the parameter of the Rayleigh distribution I Smallsample theory for censored order statisticsrdquo IEEE Transactionson Reliability vol R-22 no 1 pp 27ndash34 1973
[12] A A Soliman and F M Al-Aboud ldquoBayesian inference usingrecord values from Rayleigh model with applicationrdquo EuropeanJournal of Operational Research vol 185 no 2 pp 659ndash6722008
[13] StatSoft STATISTICA Software Version 7 StatSoft Tulsa Okla2005
[14] W-C Lee J-W Wu M-L Hong L-S Lin and R-L ChanldquoAssessing the lifetime performance index of Rayleigh productsbased on the Bayesian estimation under progressive type IIright censored samplesrdquo Journal of Computational and AppliedMathematics vol 235 no 6 pp 1676ndash1688 2011
[15] M Ahsanullah Record Statistics Nova Science New York NYUSA 1995
[16] B C Arnold N Balakrishnan and H N Nagaraja RecordsJohn Wiley amp Sons New York NY USA 1998
[17] P W Zehna ldquoInvariance of maximum likelihood estimatorsrdquoAnnals of Mathematical Statistics vol 37 p 744 1966
[18] A Zellner ldquoBayesian estimation and prediction using asymmet-ric loss functionsrdquo Journal of the American Statistical Associa-tion vol 81 no 394 pp 446ndash451 1986
[19] S-Y Huang ldquoEmpirical Bayes testing procedures in somenonexponential families using asymmetric linex loss functionrdquoJournal of Statistical Planning and Inference vol 46 no 3 pp293ndash309 1995
[20] R Calabria andG Pulcini ldquoPoint estimation under asymmetricloss functions for left-truncated exponential samplesrdquo Commu-nications in StatisticsmdashTheory and Methods vol 25 no 3 pp585ndash600 1996
[21] A T K Wan G Zou and A H Lee ldquoMinimax and Γ-minimaxestimation for the Poisson distribution under LINEX loss whenthe parameter space is restrictedrdquo Statistics and ProbabilityLetters vol 50 no 1 pp 23ndash32 2000
[22] S Anatolyev ldquoKernel estimation under linear-exponential lossrdquoEconomics Letters vol 91 no 1 pp 39ndash43 2006
[23] X Li Y Shi J Wei and J Chai ldquoEmpirical Bayes estimators ofreliability performances using LINEX loss under progressivelytype-II censored samplesrdquo Mathematics and Computers inSimulation vol 73 no 5 pp 320ndash326 2007
[24] D K Dyer and P S L Liu ldquoOn comparison of estimators in ageneralized life modelrdquo Microelectronics Reliability vol 32 no1-2 pp 207ndash221 1992
[25] A A Soliman ldquoEstimation of parameters of life from progres-sively censored data using Burr-XII modelrdquo IEEE Transactionson Reliability vol 54 no 1 pp 34ndash42 2005
[26] A A Soliman and G R Elkahlout ldquoBayes estimation of thelogistic distribution based on progressively censored samplesrdquoJournal of Applied Statistical Science vol 14 no 3-4 pp 281ndash2932005
[27] C Caroni ldquoThe correct ldquoball bearingsrdquo datardquo Lifetime DataAnalysis vol 8 no 4 pp 395ndash399 2002
[28] M Z Raqab and M T Madi ldquoBayesian prediction of the totaltime on test using doubly censored Rayleigh datardquo Journal ofStatistical Computation and Simulation vol 72 no 10 pp 781ndash789 2002
[29] W C Lee ldquoStatistical testing for assessing lifetime performanceIndex of the Rayleigh lifetime productsrdquo Journal of the ChineseInstitute of Industrial Engineers vol 25 pp 433ndash445 2008
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Journal of Applied Mathematics 13
businesses for the Pareto lifetime model with the right type IIcensored samplerdquo Applied Mathematics and Computation vol184 no 2 pp 336ndash350 2007
[4] C W Hong J W Wu and C H Cheng ldquoComputational pro-cedure of performance assessment of lifetime index of Paretolifetime businesses based on confidence intervalrdquo Applied SoftComputing vol 8 no 1 pp 698ndash705 2008
[5] J-W Wu H-M Lee and C-L Lei ldquoComputational testingalgorithmic procedure of assessment for lifetime performanceindex of products with two-parameter exponential distribu-tionrdquo Applied Mathematics and Computation vol 190 no 1 pp116ndash125 2007
[6] W-C Lee J-W Wu and C-W Hong ldquoAssessing the lifetimeperformance index of products with the exponential distribu-tion under progressively type II right censored samplesrdquo Journalof Computational and Applied Mathematics vol 231 no 2 pp648ndash656 2009
[7] W C Lee J W Wu and C L Lei ldquoEvaluating the lifetimeperformance index for the exponential lifetime productsrdquoApplied Mathematical Modelling vol 34 no 5 pp 1217ndash12242010
[8] K N Chandler ldquoThe distribution and frequency of recordvaluesrdquo Journal of the Royal Statistical Society B vol 14 pp 220ndash228 1952
[9] S K Bhattacharya and R K Tyagi ldquoBayesian survival analysisbased on the rayleigh modelrdquo Trabajos de Estadistica vol 5 no1 pp 81ndash92 1990
[10] A D Cliff and J K Ord ldquoModel building and the analysis ofspatial pattern in human geography (with discussion)rdquo Journalof the Royal Statistical Society B vol 37 no 3 pp 297ndash348 1975
[11] D D Dyer and C W Whisenand ldquoBest linear unbiasedestimator of the parameter of the Rayleigh distribution I Smallsample theory for censored order statisticsrdquo IEEE Transactionson Reliability vol R-22 no 1 pp 27ndash34 1973
[12] A A Soliman and F M Al-Aboud ldquoBayesian inference usingrecord values from Rayleigh model with applicationrdquo EuropeanJournal of Operational Research vol 185 no 2 pp 659ndash6722008
[13] StatSoft STATISTICA Software Version 7 StatSoft Tulsa Okla2005
[14] W-C Lee J-W Wu M-L Hong L-S Lin and R-L ChanldquoAssessing the lifetime performance index of Rayleigh productsbased on the Bayesian estimation under progressive type IIright censored samplesrdquo Journal of Computational and AppliedMathematics vol 235 no 6 pp 1676ndash1688 2011
[15] M Ahsanullah Record Statistics Nova Science New York NYUSA 1995
[16] B C Arnold N Balakrishnan and H N Nagaraja RecordsJohn Wiley amp Sons New York NY USA 1998
[17] P W Zehna ldquoInvariance of maximum likelihood estimatorsrdquoAnnals of Mathematical Statistics vol 37 p 744 1966
[18] A Zellner ldquoBayesian estimation and prediction using asymmet-ric loss functionsrdquo Journal of the American Statistical Associa-tion vol 81 no 394 pp 446ndash451 1986
[19] S-Y Huang ldquoEmpirical Bayes testing procedures in somenonexponential families using asymmetric linex loss functionrdquoJournal of Statistical Planning and Inference vol 46 no 3 pp293ndash309 1995
[20] R Calabria andG Pulcini ldquoPoint estimation under asymmetricloss functions for left-truncated exponential samplesrdquo Commu-nications in StatisticsmdashTheory and Methods vol 25 no 3 pp585ndash600 1996
[21] A T K Wan G Zou and A H Lee ldquoMinimax and Γ-minimaxestimation for the Poisson distribution under LINEX loss whenthe parameter space is restrictedrdquo Statistics and ProbabilityLetters vol 50 no 1 pp 23ndash32 2000
[22] S Anatolyev ldquoKernel estimation under linear-exponential lossrdquoEconomics Letters vol 91 no 1 pp 39ndash43 2006
[23] X Li Y Shi J Wei and J Chai ldquoEmpirical Bayes estimators ofreliability performances using LINEX loss under progressivelytype-II censored samplesrdquo Mathematics and Computers inSimulation vol 73 no 5 pp 320ndash326 2007
[24] D K Dyer and P S L Liu ldquoOn comparison of estimators in ageneralized life modelrdquo Microelectronics Reliability vol 32 no1-2 pp 207ndash221 1992
[25] A A Soliman ldquoEstimation of parameters of life from progres-sively censored data using Burr-XII modelrdquo IEEE Transactionson Reliability vol 54 no 1 pp 34ndash42 2005
[26] A A Soliman and G R Elkahlout ldquoBayes estimation of thelogistic distribution based on progressively censored samplesrdquoJournal of Applied Statistical Science vol 14 no 3-4 pp 281ndash2932005
[27] C Caroni ldquoThe correct ldquoball bearingsrdquo datardquo Lifetime DataAnalysis vol 8 no 4 pp 395ndash399 2002
[28] M Z Raqab and M T Madi ldquoBayesian prediction of the totaltime on test using doubly censored Rayleigh datardquo Journal ofStatistical Computation and Simulation vol 72 no 10 pp 781ndash789 2002
[29] W C Lee ldquoStatistical testing for assessing lifetime performanceIndex of the Rayleigh lifetime productsrdquo Journal of the ChineseInstitute of Industrial Engineers vol 25 pp 433ndash445 2008
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of