103 CHAPTER – 4 ACCEPTANCE SAMPLING PLANS 4.1 Introduction Acceptance sampling is concerned with inspection and decision making regarding products, one of the oldest aspects of quality assurance. In the 1930‟s and 1940‟s, acceptance sampling was one of the major components of the field of statistical quality control and was used primarily for incoming or receiving inspection. A typical application of acceptance sampling is as follows. A company receives a shipment of product from a vendor. This product is often a component or raw material used in the company‟s manufacturing process. A sample is taken from the lot and some quality characteristic of the units in the sample is inspected. On the basis of the information in this sample, a decision is made regarding lot disposition. Usually, this decision is either to accept or to reject the lot. Sometimes we refer to this decision as lot sentencing. Accepted lots are put into production; rejected lots may be returned to the vendor or may be subjected to some other lot disposition action. While it is customary to think of acceptance sampling as a receiving inspection activity, there are other uses of sampling methods. For example, frequently a manufacturer will sample and inspect its own product at various stages of production. Lots that are accepted are sent forward for further processing, while rejected lots may be reworked or scrapped.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
103
CHAPTER – 4
ACCEPTANCE SAMPLING PLANS
4.1 Introduction
Acceptance sampling is concerned with inspection and
decision making regarding products, one of the oldest aspects of quality
assurance. In the 1930‟s and 1940‟s, acceptance sampling was one of the
major components of the field of statistical quality control and was used
primarily for incoming or receiving inspection.
A typical application of acceptance sampling is as follows. A
company receives a shipment of product from a vendor. This product is
often a component or raw material used in the company‟s manufacturing
process. A sample is taken from the lot and some quality characteristic of
the units in the sample is inspected. On the basis of the information in this
sample, a decision is made regarding lot disposition. Usually, this
decision is either to accept or to reject the lot. Sometimes we refer to this
decision as lot sentencing. Accepted lots are put into production; rejected
lots may be returned to the vendor or may be subjected to some other lot
disposition action. While it is customary to think of acceptance sampling
as a receiving inspection activity, there are other uses of sampling
methods. For example, frequently a manufacturer will sample and inspect
its own product at various stages of production. Lots that are accepted are
sent forward for further processing, while rejected lots may be reworked
or scrapped.
104
It is the purpose of acceptance sampling to sentence lots, not to
estimate the lot quality. Most acceptance sampling plans are not designed
for estimation purposes.
Acceptance sampling plans do not provide any direct form of
quality control. Acceptance sampling simply accepts / rejects lots. Even if
all lots are of the same quality, sampling plan will accept some lots and
reject others, the accepted lots being no better than the rejected ones.
Process controls are used to control and systematically improve quality,
but acceptance sampling is not. The most effective use of acceptance
sampling is not to “inspect quality into the product” but rather as an audit
tool to ensure that the output of a process conforms to requirements.
We generally use 100% inspection in situations where the
component is extremely critical and passing any defectives would result
in an unacceptably high failure cost at subsequent stages.
Acceptance sampling is most likely to be useful in the following
situations:
1. When testing is destructive.
2. When the cost of 100% inspection is extremely high.
3. When 100% inspection is not technologically feasible or
would require so much calendar time that production
scheduling would be seriously impacted.
4. When there are many items to be inspected and the
inspection error rate is sufficiently high that 100% inspection
might cause a higher percentage of defective units to be
passed than would occur with the use of a sampling plan.
105
5. When the vendor has an excellent quality history and some
reduction in inspection from 100% is desired, but the
vendor‟s process capability is sufficiently low as to make no
inspection an unsatisfactory alternative.
There are risks of accepting “bad” lots and rejecting “good” lots.
Less information is usually generated about the product or about the
process that manufactured the product.
Acceptance sampling is a “middle ground” between the extremes
of 100% inspection and no inspection. If often provides a methodology
for moving between these extremes as sufficient information is obtained
on the control of the manufacturing process that produces the product.
While there is no direct control of quality in the application of an
acceptance sampling plan to an isolated lot, when that plan is applied to a
stream of lots from a vendor, it becomes a means of providing protection
for both the producer of the lot and the consumer. It also provides for an
accumulation of quality history regarding the process that produces the
lot, and it may provide feedback that is useful in process control, such as
determining when process controls at the vendor‟s plant are not adequate.
Finally, it may place economic or psychological pressure on the vendor to
improve the production process.
Acceptance sampling plans can be classified as variables and
attributes. Variables, of course are quality characteristics that are
measured in numerical scale. Attributes are quality characteristics that are
expressed on a “go, no-go” basis. The primary disadvantage of variables
sampling plans is that the distribution of the quality characteristic must be
known. In most of the standard variable acceptance sampling plans
assume that the distribution of the quality characteristic is normal. If the
106
distribution of the quality characteristic is not normal, and a plan based on
the normal assumption is employed. Serious departures from the
advertised risks of accepting or rejecting lots of given quality may be
experienced. Another disadvantage of variable sampling plan is that a
separate sampling plan must be employed for each quality characteristic
that is being inspected. For example, if an item is inspected with respect
to four quality characteristics, it is necessary to have four separate
variables inspection sampling plans. If this same product were being
inspected under attribute sampling, one attribute sampling plan could be
employed. Finally it is possible that the use of variable sampling plans
will lead to rejection of a lot even though the actual sample inspected
does not contain any defective items.
There are two general types of variable sampling procedures, (i)
plans that control the lot or process fraction defective (or non
conforming), and (ii) plans that control a lot or process parameter.
Consider a variable sampling plan to control the lot or process fraction
non-conforming. Since the quality characteristic is a variable, there will
exist either a lower specification limit (LSL), an upper specification limit
(USL) or both, that define the acceptable values of this parameter.
The fraction defective in the lot is a function of the lot or process
parameters. Variable sampling plans can also be used to give assurance
regarding the average quality of a material instead of the fraction
defective. The general approach employed in this type of variable
sampling is statistical hypothesis testing.
In scaled densities a null hypothesis about scale parameter such as
“the scale parameter is greater than or equal to a specified value‟‟ is
equivalent to saying that the “average life of a product governed by the
107
given scaled density exceeds a specified value”. Acceptance of this
hypothesis by a test procedure means that the sample life times used for
testing indicate that the lot from which the sample is drawn is a good lot.
Similarly rejection of the hypothesis implies that the lot is a bad lot. Also
instead of mean an equally competent population measure of central
tendency is median especially for skewed models (Balakrishnan et al.
(2007)). Accordingly, acceptance of a null hypothesis about population
median „the median life of a product governed by a scaled density
exceeds a specified value‟ may also indicate that the lot from which the
sample is drawn is a good lot. As an extension, because median is the
fiftieth percentile, a similar test procedure for testing a null hypothesis
can be considered for any general percentile of the model also. This is
parallel between the testing of hypothesis in scaled densities and
sampling plans is the basis of this chapter.
Acceptance sampling plans in statistical quality control concern
with accepting or rejecting a submitted lot of a large size of products on
the basis of the quality of products inspected in a sample taken from the
lot. If the quality of the product that is inspected is the life time of the
product that is put for testing, after the completion of sampling
inspection what we have is a sample of life times of the sampled
products. If a decision to accept or reject the lot subject to the risks
associated with the two types of errors (rejecting a good lot/ accepting a
bad lot) is possible, such a procedure may be termed as „Acceptance
sampling based on life tests‟ or „Reliability test plans‟. Such a procedure
obviously requires the specification of the probability model governing
the life of the products. Exponential distribution-the CFR model is the
central distribution in reliability studies. Epstein (1954) developed
reliability test plans for exponential distribution. Gupta and Groll (1961)
108
constructed sampling plans similar to those of Epstein (1954) based on
Gamma distribution. Sampling plans similar to those of Gupta and Groll
(1961) are developed by Kantam and Rosaiah (1998) for half-logistic
distribution and Kantam et al. (2001) for log-logistic distribution,
Rosaiah and Kantam (2005) for the inverse Rayleigh distribution and
Srinivasa Rao et al. (2009b) for Marshall-Olkin extended Lomax
distribution. Sampling plans in a new approach for log-logistic
distribution are suggested by Kantam et al. (2006a) and Srinivasa Rao
et al. (2009a) for Marshall-Olkin extended Lomax distribution
All these authors considered the design of acceptance sampling
plans based on the population mean under a truncated life test. Whereas
Lio et al. (2010) have considered acceptance sampling plans from
truncated life tests based on the Birnbaum-Saunders distribution for
percentiles and they proposed that the acceptance sampling plans based
on mean may not satisfy the requirement of engineering on the specific
percentile of strength or breaking stress. When the quality of a specified
low percentile is concerned, the acceptance sampling plans based on the
population mean could pass a lot which has the low percentile below the
required standard of consumers. Furthermore, a small decrease in the
mean with a simultaneous small increase in the variance can result in a
significant downward shift in small percentiles of interest. This means
that a lot of products could be accepted due to a small decrease in the
mean life after inspection. But the material strengths of products are
deteriorated significantly and may not meet the consumer‟s expectation.
Therefore, engineers pay more attention to the percentiles of lifetimes
than the mean life in life testing applications. Moreover, most of the
employed life distributions are not symmetric. In viewing Marshall and
Olkin (2007), the mean life may not be adequate to describe the central
109
tendency of the distribution. This reduces the feasibility of acceptance
sampling plans if they are developed based on the mean life of products.
Actually, percentiles provide more information regarding a life
distribution than the mean life does. When the life distribution is
symmetric, the 50th
percentile or the median is equivalent to the mean
life. Hence, developing acceptance sampling plans based on percentiles of
a life distribution can be treated as a generalization of developing
acceptance sampling plans based on the mean life of items. Balakrishnan
et al. (2007) proposed the acceptance sampling plans could be used for
the quantiles and derived the formulae whereas Lio et al. (2010)
developed for the acceptance sampling plans for any other percentiles of
the Birnbaum-Saunders (BS) model. They have developed the acceptance
sampling plans for percentile by replace the scale parameter by the
100qth percentile in the BS distribution function. Srinivasa Rao and
Kantam (2010) are developed acceptance sampling plans from truncated
life tests based on the log-logistic distribution for percentiles.
In this chapter, we develop acceptance sampling plans based on
average (mean) for the inverse Rayleigh distribution under a truncated
life test, along with operating characteristic and relevant tables, example
based on real life data sets are provided in Section 4.2. The proposed
sampling plans based on percentiles, along with operating characteristic
function and examples based on real life data sets are provided for the
illustration in Section 4.3. A comparative study of the two approaches is
given in Section 4.4. The discussion and some conclusions are made in
Section 4.5.
110
4.2 Acceptance sampling plans - Averages
We assume that the life time of a product follows an inverse
Rayleigh distribution with scale parameter . A common practice in life
testing is to terminate a life test by a pre-determined time t and note the
number of failures (assuming that a failure is well-defined). One of the
objectives of these experiments is to set a lower confidence limit on the
average life. It is then desired to establish a specified average life with a
given probability of at least P*. The decision to accept the specified
average life occurs if and only if the number of observed failures at the
end of the fixed time t does not exceed a given number c- called the
acceptance number. The test may get terminated before the time t is
reached when the number of failures exceeds c in which case, the
decision is to reject the lot. For such a truncated life test and the
associated decision rule, we are interested in obtaining the smallest
sample sizes necessary to achieve the objective.
A sampling plan consists of
(i) The number of units n on test,
(ii) The acceptance number c,
(iii) The maximum test duration t and
(iv) The ratio t/o, where o is the specified average life
The consumer‟s risk, i.e., the probability of accepting a bad lot (the
one for which the true average life is below the specified life o) not to
exceed 1-P*, so that P
* is a minimum confidence level with which a lot of
true average life below o is rejected, by the sampling plan. For a fixed
P* our sampling plan is characterized by (n, c, t/o). Here we consider
sufficiently large lots so that the binomial distribution can be applied. The
111
problem is, for given values of P* (0<P
*<1), o and c the smallest
positive integer n is to be determined such that
0
(1 ) 1c
i n i
i
np p P
i
(4.2.1)
where p=F(t;o) is given by (4.2.1 ) indicates the failure probabilities
before time t, which depends only on the ratio t/, it is sufficient to
specify this ratio for designing the experiment.
If the number of observed failures before t is less than or equal to c,
from inequality (4.2.1) we obtain
F(t ; ) F(t ; o) o (4.2.2)
The minimum values of n satisfying the inequality (4.2.1) are
obtained and given in Table (4.2.1) for P*=0.75, 0.90, 0.95, 0.99 ;
t/o=1.0, 1.25, 1.5, 1.75, 2.0, 2.25, 2.5, 3.0, 3.5, 4.0 and c = 0, 1, 2, …,
10.
If p=F(t; ) is small and n is large (as is true in some cases of our
present work), the binomial probability may be approximated by Poisson
probability with parameter =np so that the left hand side of (4.2.1) can
be written as
1!0
ic eP
ii
(4.2.3)
where =nF(t;). The minimum values of n satisfying (4.2.3) are
obtained for the same combination of P*, t/o and c values as those used
for (4.2.1). The results are given in Table 4.2.2
112
Operating characteristic function
In the problem of choosing an acceptance sampling plan, interest
centers on how it will perform in actual practice not merely for one value
of a specified lot quality parameter (specified average in the present case)
but for all possible values that might be encountered. In other words, the
probabilities of a lot being accepted should have clear relation with
changes in lot quality parameter. In the present situation any downward
shift in the average life of the product should reveal a significant fall in
the probability of acceptance of the lot. This is best shown by
considering the graph of acceptance probability of the sampling plan
against the lot quality parameter. Such a graph is called operating
characteristic (O.C) curve. There are two distinct types of operating
characteristic curves called type A and type B according as sampling is
from a finite universe or infinite universe. Our sampling plans will have
type B operating characteristic curves.
The operating characteristic function of the sampling plan (n, c,
t/o) gives the probability L(p) of accepting the lot with :
L(p)=0
(1 )c
i n i
i
np p
i
(4.2.4)
where p=F(t;) is considered as a function of , i.e., the lot quality
parameter. It can be seen that operating characteristic is an increasing
function of . For given P*, t/o , the choice of c and n will be made on
the basis of operating characteristics. Values of operating characteristics
as a function of /o for a few sampling plans are given in Table 4.2.3
when the plan is based on binomial probability. Similar values are given
113
in Table 4.2.2 for a sampling plan based on Poisson probabilities are not
much different from that of Table 4.2.1, and hence, O.C values for these
are not presented.
Scope of population average relative to a producer’s risk
In life test data required for sampling plans of quality control, we
know that lots having the average life above a specification are termed as
good lots. It may be noted that the quality parameter considered here is
population average life of the product a case of “the more, the better”.
That is, all lots with products having average life beyond the specified life
(σ>σ0) are regarded as good lots. On the application of sampling plan if a
good lot is not accepted sometimes, it results in a wrong decision also
called an error in classical testing of statistical hypothesis. The probability
of rejecting a good lot on the application of an acceptance sampling is
called producer‟s risk. It may be recalled that the sampling plans
presented by us in section 4.2 are constructed for a given consumer‟s
risk - probability of accepting a bad lot. Therefore, in this section we
attempt to investigate how far the specified and the unknown population
average fluctuate without disturbing producer‟s risk. In other words, let
the producer‟s risk be 0.05. Let 0 be the ratio of specified average to
unknown average. We attempt to find the range of values of 0 which
will ensure a producers risk not beyond 0.05 if a specified sampling plan
is adopted. We know that the distribution function of the population
distribution can be written as a function of 0 .
It should be noted that the probability p may be obtained as a
function of 0 , as
p=F(t/)=F[(t/o)(o/)]. (4.2.5)
114
where p=F (t;).
The value 0 is the smallest number for which the following
inequality
0
(1 ) 0.95c
i n i
i
np p
i
(4.2.6)
For a given sampling plan (n, c, t/o) on the basis of binomial
probabilities and specified confidence level P*, the minimum values of
0 satisfying the inequality (4.2.6) are given in Table 4.2.4. Similar
values for sampling plan constructed using Poisson probabilities are not
much different from binomial probabilities and hence, these are not
presented.
Illustration of the Tables and Example for Test plan
Assume that the life distribution is an inverse Rayleigh distribution
and the experimenter is interested in showing that the true unknown
average life is at least 1000 hours. Let the consumer‟s risk be set to
1-P*=0.25. It is desired to stop the experiment at t=1000 hours. Then for
an acceptance number c=2, the required n is the entry in Table 4.2.1
corresponding to the values of 1-P*=0.25, t/o=1.0 and c=2. This number
is n=10. Thus n=10 units have to be put on test. If during 1000 hours, no
more than 2 failures out of 10 units are observed, then the experimenter
can assert with a confidence level of P*=0.75 that the average life is at
least 1000 hours. If the Poisson approximation to binomial probability is
used, the value of n=11 is obtained for the same situation from Table
4.2.2.
For the sampling plan (n=10, c=2, t/o=1.0) under inverse Rayleigh
model, the operating characteristic values from Table 4.2.3 are
115
/o : 2 4 6 8 10 12
L(p) : 0.999 1.000 1.000 1.000 1.000 1.000
This shows that, if the true mean life is twice the specified mean
life ( 0 =2) the producer‟s risk is approximately 0.001.
From Table 4.2.4, we can get the value of 0 for various choices
of c, t/o in order that the producer‟s risk may not exceed 0.05. Thus, in
the above example we obtain the values of 0 = 1.81. That is, the
product should have an average life of 1.81 times the specified average
life 1000 hours in order that under the above acceptance sampling plan
(10, 2, 1.0) the product be accepted with probability of at least 0.95. The
actual average life necessary to transship 95 percent of the lots is
provided by Table 4.2.4.
Example: Consider the following ordered failure times of the release of
a software given in terms of hours from the starting of the execution of
the software denoting the times at which the failure of the software is
experienced (Wood, 1996). This data can be regarded as an ordered
sample of size 10 with observations (Xi , i=1, 2, 3, . . . , 10) : 519, 968,
1430, 1893, 2490, 3058, 3625, 4422, 5218, 5823.
Let the specified average life be 1000 hrs and the testing time be
1250 hrs, this leads to ratio of t/o=1.25 with corresponding n and c as
10, 2 from Table 4.2.1 for P*=0.95. Therefore, the sampling plan for the
above sample data is (n=10, c=2, t/o=1.25). Based on the observations,
we have to decide whether to accept the product or reject it. We accept
the product only, if the number of failures before 1250 hrs is less than or
equal to 2. However, the confidence level is assured by the sampling
plan only if the given life times follow an inverse Rayleigh distribution,
116
we have compared with the sample quantiles and the corresponding
population quantiles and found a satisfactory agreement .Thus the
adoption of the decision rule of the sampling plan seems to be justified.
We see that in the sample of 10 failures there are 2 failures at 519 and
968 hrs before 1250 hrs, therefore we accept the product.
4.3 Acceptance sampling plans - Percentiles
Assume that the lifetime of a product follows a inverse Rayleigh
distribution with scale parameter . For 0 1q , the 100qth percentile (or
the qth quantile) is given by
1/2
lnqt q
(4.3.1)
The qt is increases as q increases. Let 1/ 2
ln q
. Then, equation
(4.3.1) implies that
qt . (4.3.2)
To develop acceptance sampling plans for the inverse Rayleigh
percentiles, the scale parameter in the inverse Rayleigh cdf is replaced
by equation (4.3.2) and the inverse Rayleigh cdf is rewritten as
2
( ) /( ) ; 0qt t
F t e t
.
Lettingqt t , F(t) can be rewritten emphasizing its dependence on
as
21( ; ) ; 0F t e t
.
Taking partial derivative with respect to , we have
117
21
3
( ; ) 2; 0
F te t
.
A common practice in life testing is to terminate the life test by a
pre-determined time t, the probability of rejecting a bad lot be at least P ,
and the maximum number of allowable bad items to accept the lot be c.
The acceptance sampling plan for percentiles under a truncated life test is
to set up the minimum sample size n for the given acceptance number c
such that the consumer‟s risk, the probability of accepting a bad lot, does
not exceed 1-P . A bad lot means that the true 100qth
percentile, qt , is
below the specified percentile, 0
qt . Thus, the probability P is a confidence
level in the sense that the chance of rejecting a bad lot with 0
q qt t is at
least equal to P . Therefore, for a given P , the proposed acceptance
sampling plan can be characterized by the triplet 0( , , )qn c t t .
Minimum sample size
For a fixed P , our sampling plan is characterized by 0( , , )qn c t t .
Here we consider sufficiently large sized lots so that the binomial
distribution can be applied. The problem is to determine for given values
of P (0 <P <1), 0
qt and c, the smallest positive integer n required to
assert that 0
q qt t must satisfy
0 0
0
1 1c
n in i
i
i
p p P
, (4.3.3)
where 0( ; )p F t is the probability of a failure during the time t given a
specified 100qth percentile of lifetime 0
qt and depends only on 0
0 qt t ,
118
since ( ; ) 0, ( ; )F t F t is a non-decreasing function of .
Accordingly, we have
0 0( , ) ( , )F t F t ,
or equivalently,
00( , ) ( , ) q qF t F t t t .
The smallest sample size n satisfying the inequality (4.3.3) can be
obtained for any given q, 0
qt t , P . Whereas, the smallest sample size n
calculation in Section 4.2 only needs input values for 0t and P .
Hence, the proposed process to find the smallest sample size in this case
is the same as the procedure provided by in Section 4.2 or the inverse
Rayleigh model except in place of 0t replace by 0
qt t at q. To save
space, only the results of small sample sizes for q=0.10, 0.25. 0.50, 0.75,
0.90; 0
qt t =0.7, 0.9, 1.0, 1.5, 2.0, 2.5, 3.0, 3.5; P =0.75, 0.90, 0.95, 0.99;
c = 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 are reported in Tables 4.3.1- 4.3.5. These
tables reveal that as percentile values increase the smallest sample size
require are decreases for all combinations of parameters of, 0
qt t , P
and c.
If 0( ; )p F t is small and n is large the binomial probability may
be approximated by Poisson probability with parameter λ = np so that the
left side of (4.3.3) can be written as
0
1 P!
ic
i
ei
, (4.3.4)
119
where λ = n. 0( ; )F t . The minimum values of n satisfying (4.3.4)
are obtained for the same combination of q, 0
qt t and P values as those
used for (4.3.3). The results are reported in Tables 4.3.6 to 4.3.10.These
tables reveal that as percentile values increase the smallest sample size
require are decreases for all combinations of parameters of 0
qt t , P
and c. Also as compared with binomial approach, the Poisson approach
gives the smaller sample size.
Operating characteristic of the sampling plan 0( , , )qn c t t
The operating characteristic (OC) function of the sampling plan
0( , , )qn c t t is the probability of accepting a lot. It is given as
0
( ) 1c
n in i
i
i
L p p p
, (4.3.5)
where ( ; )p F t . It should be noticed that ( ; )F t can be represented as
a function of qt t . Therefore,
0
1( )
q q
tp F
t d where 0
q q qd t t . Using
equation (4.3.5), the OC values and OC curves can be obtained for any
sampling plan 0( , , )qn c t t . To save space, we present Tables 4.3.11–
4.3.15 to show the OC values for the sampling plan 0
0.1( , 5, )n c t t .
Figures 4.3.1 – 4.3.5 shows the OC curves for the sampling plan
0
0.1( , , )n c t t with P =0.90 for 0 0.8 , for q=0.10, 0.25. 0.50, 0.75, 0.90,
when c= 0,1,2,3,4,5,6,7,8,9,10. These OC curves show that as percentile
increases the OC values are decreases for all given combinations.
120
Figure 4.3.1: OC curves for c = 0,1,2,3,4,5,6,7,8,9,10, respectively under
P =0.90, 0 0.8 based on the 10th percentile, 0.1d d , of inverse
Rayleigh distribution.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.00 1.25 1.50 1.75 2.00 2.25 2.50 2.75
L(P
)
d
Operating Characterstic Curve
Figure 4.3.2: OC curves for c = 0,1,2,3,4,5,6,7,8,9,10, respectively under
P =0.90, 0 0.8 based on the 25th percentile, 0.25d d , of inverse
Rayleigh distribution.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.00 1.25 1.50 1.75 2.00 2.25 2.50 2.75
L(p
)
d
Operating Characterstic Curve
121
Figure 4.3.3: OC curves for c = 0,1,2,3,4,5,6,7,8,9,10, respectively under
P =0.90, 0 0.8 based on the 50th percentile, 0.50d d , of inverse
Rayleigh distribution.
Figure 4.3.4: OC curves for c = 0,1,2,3,4,5,6,7,8,9,10, respectively under
P =0.90, 0 0.8 based on the 75th percentile, 0.75d d , of inverse
Rayleigh distribution.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.00 1.25 1.50 1.75 2.00 2.25 2.50 2.75
L(P
)
d
Operating Characteristic Curve
122
Figure 4.3.5: OC curves for c = 0,1,2,3,4,5,6,7,8,9,10, respectively under
P =0.90, 0 0.8 based on the 90th percentile, 0.90d d , of inverse
Rayleigh distribution.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.00 1.25 1.50 1.75 2.00 2.25 2.50 2.75
L(P
)
d
Operating Characteristic Curve
Producer’s Risk
The producer‟s risk is defined as the probability of rejecting the lot
when 0
q qt t . For a given value of the producer‟s risk, say , we are
interested in knowing the value of qd to ensure the producer‟s risk is less
than or equal to if a sampling plan 0( , , )qn c t t is developed at a
specified confidence level p . Thus, one needs to find the smallest value
qd according to equation (4.3.5) as
0
1 1c
n in i
i
i
p p
, (4.3.6)
123
where 0
1( )
q q
tp F
t d , 0
q q qd t t . To save space, based on sampling plans
0( , , )qn c t t the minimum ratios of 0.1 0.25 0.5 0.75 0.9, , , andd d d d d for the
acceptability of a lot at the producer‟s risk of =0.05 are presented in
Tables 4.3.16 to 4.3.20.
Illustrative Examples
In this section, we consider two examples with real data sets are
given to illustrate the proposed acceptance sampling plans. The first data
set is of the data given arisen in tests on endurance of deep groove ball
bearings (Lawless, 1982, p.228). The data are the number of million
revolutions before failure for each of the 23 ball bearings in life test and