Professional Documents
Culture Documents
Abstract
To conduct Acceptance Sampling, companies often use plans that are determined by easy-touse standards. However, these standards do not take quality costs directly into account. Motivated
by the case of a Greek company, which uses the Greek equivalent to the ISO 2859 (1974) for the
quality control of its incoming raw materials, this paper aims at evaluating the single-sampling
plans recommended by the latest update of ISO 2859, from an economic point of view. The
evaluation shows that the use of standards rarely leads to satisfactory economic results. Therefore,
simple rules for the economical use of the standard are provided.
Key words: Statistical Quality Control; Acceptance Sampling; Sampling Plan; Economic Design;
Quality Cost
1.
Introduction
Over the last decades quality issues have held the attention of both industry and research
communities because of their impact on cost and, therefore, on profit. Quality costs include any
cost that results from the fact that systems, processes, products and services are imperfect. More
specifically, it consists of prevention costs, appraisal costs, internal and external failure costs, and it
varies between 4% and 40% of the sales of a company (Montgomery, 2001). Consequently, the
analysis and estimation of quality cost are of great importance. Usually, all quality cost categories
can be expressed as functions of the actual quality of products (e.g. fraction nonconforming of a
lot), therefore, it is possible to derive the economically optimum quality of products, as well as the
optimum quality control policy, i.e., the control policy that minimizes the expected quality-related
costs.
Statistical Quality Control (SQC) aims at monitoring and improving the quality of products
produced by a process and consists of three areas/types: Design of Experiment, Statistical Process
Control and Acceptance Sampling. Modern quality assurance systems usually place less emphasis
on the latter, while they attempt to focus their efforts on the other two types of SQC. However,
there are still numerous companies all over the world where the evolution of SQC techniques is
limited and, thus, they use mainly Acceptance Sampling in order to monitor the quality of either
incoming materials or final products. For example, in most small to medium-sized Greek
companies, the typical SQC procedure is lot-by-lot Acceptance Sampling, single-sampling by
attributes in particular. In our research we focus on this type of SQC.
In general, single-sampling by attributes is carried out as follows: from a lot of N units, a
sample of size n is taken and every item of the sample is checked and characterized as conforming
or nonconforming based on an attribute-type quality characteristic, e.g., color, appearance etc. The
lot may be accepted or rejected depending on the outcome of the control and the acceptance rejection criterion, i.e., the maximum allowable number of nonconforming items in the sample
(acceptance number c).
In order to monitor the quality of acquired or produced lots, companies often use sampling
plans (n,c) that are recommended by specific standards, which are simple and easy to use. Two
recently released technical reports that provide guidance for Acceptance Sampling of products in
lots are ISO/TR 8550-1 (2007) and ISO/TR 8550-2 (2007). However, the most popular
international standard is ISO 2859-10 (2006). In fact, this standard is a review of the ISO 2859
series of standards and it does not include any tables of sampling plans. The latter can be found in
any of the five standards - members of the ISO 2859-10 (2006) family of standards, namely ISO
2859-1 (1999), ISO 2859-2 (1985), ISO 2859-3 (2005), ISO 2859-4 (2002) and ISO 2859-5 (2005),
which are suitable for various Acceptance Sampling cases. Although in our paper we use the
sampling plans recommended by ISO 2859-1 (1999), in the analysis of the paper hereafter, we will
use the general term ISO 2859.
In Greece, Acceptance Sampling is usually conducted by using the ELOT 398.0 (1982)
standard and its addendum ELOT 398.1 (1982). Together, ELOT 398.0 and 398.1 constitute the
Greek equivalent to the ISO 2859 (1974), namely, to an outdated version of the ISO 2859 family of
standards. Note that although the ISO 2859 standard has been updated many times since 1974, the
respective Greek standards remain the same, not following the updating procedure of the
international standard. Nevertheless, despite the updates of ISO 2859, the part of the latest version
that we examine in our research has remained unchanged through years and therefore, our results
that were based on ELOT 398.0 and 398.1 standards, hold also for the newer versions of the ISO
2859 family.
The motivation for this research was given by a project that we undertook a few years ago,
that was intended to reduce the quality-related costs incurred during the Acceptance Sampling
process, using the ELOT 398.0 and 398.1 standards, of the incoming raw materials of CRYSTAL
S.A., a commercial refrigerators Greek company. The main conclusion of that project was that the
ELOT standards were leading to unacceptably high quality costs very often. Having realized that,
we have extended our research in order to come up with a simple and easy way to use either the
ELOT 398.0 and 398.1 standards or, most importantly, the ISO 2859, without overlooking their
inevitable economic impact. Consequently, this paper focuses on determining the economic
consequences of using a quality standard without taking into account the respective quality costs.
In any case, we dont downplay the ISO 2859 standard. We just present some rules that can lead to
its better use, considering that this particular standard per se is a tool for statistical quality control
with wide-ranging appeal. It should be mentioned that this effort has been embraced by a lot of
companies in Greece, mainly because it reveals some weaknesses of quality standards concerning
their economic efficiency, which the companies havent realized until now.
The implementation of single-sampling plans by attributes that are recommended by the ISO
2859, is evaluated economically for a variety of a) parameters of the control process, i.e., lot size
(N), Acceptance quality limit (AQL) etc., b) cost elements and c) quality level of lots, viz the
fraction nonconforming (p). In particular, we determine the average total quality cost
corresponding to the plans (n,c) recommended by the standard and then we compare it against the
respective cost of the optimum Acceptance Sampling plans (n*,c*). In addition, we compare the
parameters of the former plans against the optimum parameters, i.e., the plan parameters that lead
to the minimization of the average total quality cost.
The remainder of the paper is structured as follows. The following Section presents the related
literature. In Section 3 we set the notation and we formulate the average total quality cost function.
Section 4 briefly presents the case of CRYSTAL S.A. The combinations of parameters that are
examined and the extensive numerical investigation are presented in Section 5, while the final
Section of the paper summarizes the main conclusions of our research and proposes topics for
future research.
2.
Related Literature
Detailed information about SQC techniques can be found in many books, such as the one by
Montgomery (2001). In Tagaras (2001) there is a section dealing with the economic aspect of
Acceptance Sampling plans. More specialized information about Acceptance Sampling and its
economic dimension can be found in Hald (1981).
Significant reviews of papers concerning Acceptance Sampling are those of Wetherrill and
Chiu (1975) and Wall and Elshennawy (1989). Both papers make extensive references to
economically optimum sampling plans. Bai and Riew (1984) develop an economic Acceptance
Sampling plan by attributes for cases where sampling is expensive or destructive. They present a
linear cost model and they consider three decision criteria. A few years later Tagaras and Lee
(1987) develop an algorithm to compute the economically optimum single-sampling plan, using
Bayes theorem.
Even though lately Acceptance Sampling has given place to more advanced SQC techniques,
the published research on this particular SQC area and especially on the economic design of
sampling plans is still extensive. Thus, Ferrell and Chhoker (2002) have recently proposed an
economic model for the design of Acceptance Sampling plans adopting the Taguchi approach.
Gonzlez and Palomo (2003) use a Bayesian analysis in order to derive Acceptance Sampling plans
regarding the number of defects per unit of product and apply their methodology to the paper pulp
industry. The sampling plans are obtained following an economic criterion: the minimization of the
expected total cost of quality. At the same time, Cassady and Nachlas (2003) define a generic
framework for establishing three-level acceptance sampling plans, using quality value functions.
They note that there are many cases in which the quality of a product can be classified in three or
more discrete levels; for example, a food product may be classified as good, marginal or bad. Chen
et al. (2004b) present and investigate a general model of Acceptance Sampling plan for the
exponential distribution with exponentially distributed random censoring, based on Bayesian
decision theory. In order to determine the optimal sampling plan they consider a loss function,
which includes the sampling cost, the time-consuming cost and the decision loss to determine the
optimal Acceptance Sampling plan. At the same time and in a similar study Chen et al. (2004a)
develop a general Bayesian framework for designing a variable Acceptance Sampling plan with
mixed censoring. A general loss function including the three partial costs of Chen et al. (2004b) as
well as the salvage value is introduced to determine the corresponding optimal sampling plan. Stout
and Hardwick (2005) present a unified approach to the problem of response adaptive screening,
when multiple costs and constraints ought to be incorporated. In particular, they describe a costand constraint-based approach, which is suitable in Acceptance Sampling where sample products
must be tested before an entire batch is accepted. Finally, Chen (2006) modifies Pulak and AlSultans model in order to determine the optimum process mean and standard deviation under the
rectifying inspection plan with the average outgoing quality limit protection. The symmetric
quadratic quality loss function of Taguchi is adopted for evaluating the product quality.
The forerunner of the present paper is some earlier work by Nikolaidis and Nenes (2005),
regarding the project mentioned previously, which deals with the economic consequences of using
the ELOT 398.0 and 398.1 standards, for the control of some raw materials of CRYSTAL S.A.
Regarding the updates of the ISO 2859 standard, it has been updated twice since 1974. More
specifically, ISO 2859 (1974) has been revised by ISO 2859-1 (1989), which in turn has been
replaced by ISO 2859-1 (1999). The latter is currently the applied standard and, as mentioned
previously, it constitutes a part of the recently published ISO 2859-10 (2006).
3.
sampling plan using economic criteria, it is necessary to determine the economic elements of the
control process first and then to formulate an average total quality cost function. The form of this
function depends on the characteristics of the control process. The calculation or the minimization
(using either analytical or numerical methods) of that function ensues.
In the case of Acceptance Sampling by attributes, the quality cost categories mentioned
previously are modified accordingly; normally, in the average total quality cost function one can
find the sampling cost, the cost of handling nonconforming units (e.g., repair cost, returns etc.) and
the cost of nonconforming items not detected during inspection (e.g., use of nonconforming raw
materials in the production process, defamation of a company in case that nonconforming products
reach its customers etc.). The notation of the respective cost elements is the following:
ci:
inspection cost per item, which is determined by reckoning the time required to inspect a unit,
the salary of the inspector(s), the cost of the control device(s) etc.,
cr: replacement cost, i.e., purchase cost, in case a nonconforming unit is immediately replaced
upon detection at a companys expense, or repair cost per nonconforming item,
cd: cost per nonconforming item that is not detected during inspection; cd cr, where equality
holds in case a nonconforming unit is replaced or repaired with no additional cost.
The average total quality cost per lot that corresponds to the use of a single-sampling plan
(n,c) for the inspection of a lot with a fraction nonconforming p, is in general:
K n, c p [cost of accepting a lot] Pa p +[cost of rejecting a lot] 1 Pa p ,
c
where Pa p
n!
d!n d ! p 1 p
d
nd
(1)
d 0
nonconforming p.
In the case that a lot of size N is submitted to 100% inspection if rejected (rectifying
inspection), the analytical form of (1) becomes:
K n, c p nci npc r N n pc d Pa p Nci Npc r 1 Pa p .
(2)
Note that although the choice to apply 100% inspection on rejected lots is not always made in real
life situations, it is very common in practice and, thus, this paper focuses only on such cases.
Remark 1: through simple modifications of (2) it is easy to verify the observation of Hald
(1981) that in case of known and constant fraction nonconforming p the optimum sampling plan is
either 100% inspection, i.e., n = N - if p ci / cd cr - or acceptance without sampling, i.e., n = c
= 0 - if p ci / cd cr . The ratio ci / cd cr is called break-even quality level and is denoted
hereafter by pr.
Remark 2: again, through simple modifications of (2) it can be seen that the average inspection
cost in our cost function is a product of the inspection cost ci and the average total inspection (ATI),
which Montgomery (2001) calculates by n 1 Pa p N n . Specifically:
K n ,c p ci ATI npc r N n pc d Pa p Npc r 1 Pa p .
(3)
Since most of the times the fraction nonconforming p per lot is not deterministic, but is
distributed according to a probability density function (pdf) (p), the average total quality cost
K n, c is given by:
K n, c K n, c p p dp .
(4)
Remark 3: a closer look on (2) and (4) reveals that the simultaneous increase or decrease of cr
and cd by A cost units, leads to an increase or decrease respectively of K n , c by ANp cost units,
independently of the distribution of p. Note that p denotes the average fraction nonconforming of
consecutive lots, whose p is distributed according to (p). However, since the term ANp is
constant, i.e., it is not affected by the parameters of the selected sampling plan (n,c), it follows that
the optimum plan (n*,c*) doesnt change, if a simultaneous change of the same magnitude on cr
and cd takes place.
4.
using the ELOT standards. Subsequently, an indicative raw material is presented in detail; the
palette, which is the basis where the refrigerator is placed during packaging. It is very important for
every palette to carry the necessary nogs in order for the packaged refrigerators to be properly
placed into containers. In all palettes that are found during quality control to be nonconforming
(i.e., without nogs), the companys inspector puts the necessary nogs. If a palette is found to be
nonconforming during the packaging process, then the proper worker should lift up the refrigerator,
remove the palette, put the nogs appropriately and, finally, place again the refrigerator on the
palette. In case that a lot of palettes is rejected, it is submitted to 100% inspection at companys
expense. According to historical data, the usual lot size is 800 palettes, while the fraction
nonconforming per lot p is distributed uniformly between 5% and 8%, i.e., p ~ U(0.05, 0.08).
In order to conduct Acceptance Sampling on the received lots of palettes, the Quality
Assurance Department of CRYSTAL S.A. has arbitrarily chosen to use general inspection level II,
normal inspection (without ever switching to tightened or reduced inspection) and AQL = 1%
(although this was not the wiser choice for AQL). According to these choices, the sampling plan
that ELOT 398.0 and 398.1 recommend is (n,c) = (80,2).
In the case of CRYSTAL S.A., due to the simple visual nature of inspections, the inspection
cost ci was determined by simply taking into account the time to inspect each unit and the salary of
the inspector. It has been estimated that ci = 0.119 per item. The replacement cost cr (repair cost)
of a nonconforming palette was also relatively easy to be determined. Every time a palette is found
to be nonconforming during quality control, it just needs the missing nogs. Consequently, cr was
calculated taking into account the time needed for this task, the salary of the inspector who puts the
necessary nogs and the cost of nogs. It has been estimated that cr = 0.535 per nonconforming
item. Finally, the nonconforming palettes, when detected during production, are repaired at a cost
that consists of the cost of nogs and the cost of the time required putting the nogs at the palette. It
has been estimated that cd = 2.3 per nonconforming item. Note that cd takes a much greater value
than cr since it is more difficult and needs more time to repair a palette, when a refrigerator has
already been placed on it.
Based on the aforementioned parameters of the sampling process, the actual quality of lots of
palettes, the cost elements and using (4) and (2), the average quality cost of the sampling plan that
is recommended by the ELOT 398.0 and 398.1 standards is calculated through K n 80, c 2
80 0.119 80 0.535 p 800 80 2.3 p Pa p
p dp 122.09 . The minimum average
a
0.05
0.08
quality cost and the respective optimum sampling plan are 119.58 and (n*,c*) = (85,7)
respectively. They are derived by computing (4) and (2) for every possible sampling plan and by
determining the optimal average quality cost and the sampling plan leading to the latter.
Consequently, the sampling plan recommended by the ELOT standards leads to an increased
quality cost of 2.1%, which for different parameters of the sampling process, actual quality of lots
and cost elements may be a lot higher. An explanation for this variation could be attributed to the
100% inspection of rejected lots, which in general is expensive. Therefore, the optimum sampling
plan (85,7) is not so tight as the (80,2) recommended by the ELOT standards, in order not to result
frequently to 100% inspection.
Note also that for the examined raw material, p r ci / cd cr 0.119 / 2.3 0.535 0.067 ,
i.e. 6.7%, which is very close to the average value of p (6.5%), according to its distribution. In this
case, as it will be explained extensively in the following Section, the standards were expected to
provide a near-optimum sampling plan. Nevertheless, pr may be a lot different from the average
value of p and the economic loss of using the aforementioned standards is usually much higher.
Undoubtedly, the benefit from using the optimum sampling plan instead of the one
recommended by the ELOT standards is marginal for the palettes. However, considering that the
incoming raw materials of CRYSTAL S.A. are numerous made obvious to the company that the
economic benefit from using the optimum sampling plans for all its incoming materials would be
significant. In addition, it justified and crowned with success the project that was undertaken.
5.
accomplished for a variety of combinations of parameters. The set of values that are chosen, as far
as the cost elements and the pdfs are concerned, primarily aim at reflecting real cases of
Acceptance Sampling in practice. As for the values of AQL and N, they are selected with a view to
covering most sub-cases that can occur every time the standard is used in real-life applications.
More precisely:
Lot size N: 40 different values are investigated, three of each one of the 13 first classes of
the ISO 2859 standard (a small, a medium and a large N value for every class) and one of
the 14th class, covering a wide range of N values.
Acceptance quality limit AQL: 11 different values are studied, i.e., all the values between
0.065% and 6.5%, covering the majority of the possible AQL values of the standard.
Cost elements ci, cr and cd : 11 different combinations of cost elements - that lead to a
variety of values of pr - are investigated and are presented in Table 1. Note that according
Actual quality level - fraction nonconforming p: we study four pdfs for p, three uniform
ones, i.e., U(0.0005, 0.005), U(0.0015, 0.04) and U(0.01, 0.07), corresponding to lots of
very good, good - normal and bad quality, respectively, and a truncated normal pdf
between 0.15% and 4%1, which will be denoted hereafter by TN(0.0015, 0.04), for reasons
of consistency in notation.
In every case that we examine we consider the general inspection level II, since this inspection
level is the most common in practice, adding realism to our numerical investigation. In addition, it
should be mentioned that according to both ELOT standards and ISO 2859, apart from the selection
of the sampling plan, there are also rules concerning skip-lot sampling procedures or switching
procedures between normal, tightened and reduced sampling. According to what happened in the
case of CRYSTAL S.A., these rules are not taken into consideration, since they are rarely applied
in practice (at least in Greece), while they are used for adapting the sampling plans when the
quality of acquired lots changes dramatically during time, which is not the case in our study2. Note
also that in the past many practitioners and researchers have admitted that the implementation of
switching rules constitutes a very complicated - and sometimes severe - procedure to be applied in
practice (Balamurali and Kalyanasundaram, 1997, Brown and Rutemiller, 1973), and as such it is
very often ignored (Gao and Tang, 2006, Taylor, 1995, Selinger, 1995). The evaluation of such
1
In order to design this pdf, we first created a normal pdf using the same mean and variance with the
uniform pdf (0.15% - 4%) and then we truncated the areas below 0.15% and above 4%. We computed the
necessary probabilities by dividing the probabilities of the initial normal pdf with the probability
corresponding to the truncated area, i.e., between 0.15% and 4%.
Note that the quality of lots may not change dramatically during time in our study, but it is not constant
either, since it is assumed to follow a specific pdf. If the fraction nonconforming p were assumed to be
cases is not in the objectives of this study. However, the present research can be extended to
account for these procedures as well.
According to our methodology, for every N and AQL values, the sampling plan (n,c)
recommended by ISO 2859 is first determined. Then, for every combination of cost elements and
pdf examined, the average cost of using this plan, K n, c , is calculated by (4) and (2). For the
same combination of cost elements and pdf, the optimum sampling plan (n*,c*), along with the
minimum expected quality cost K n*, c * are found; this has been done by calculating (4) and (2)
for all possible ns and cs and, then, by finding the minimum K n, c . Finally, using
K n, c K n*, c * .
100%, the percentage cost penalty of using the sampling plan (n,c) of
K n*, c *
ISO 2859 instead of the optimum (n*,c*), is calculated. Table 2 presents an indicative part of the
results that constituted the basis of our numerical investigation and comparative analysis. It should
be noted that the tool that was used for all calculations and the optimization of the cost function
K n, c was various programs developed specifically for this particular study, in Microsoft
FORTRAN PowerStation 4.
[Insert Table 2 about here]
Properties of the optimum sampling plans
Regarding the optimum sampling plans (n*,c*), we have noticed that Remark 1 stands also for
a stochastic fraction nonconforming p, distributed according to (p): whenever pr << p then n* =
N - or tends to N - and c* = 0 - or tends to 0 (100% or at least very tightened inspection), while
whenever pr >> p then n* = c* = 0 (acceptance without sampling). Generally, the increase of pr
gives more and more reduced optimum sampling plans, namely plans with continuously decreasing
n* and usually increasing c*.
Since in most cases ISO 2859 allows neither 100% inspection nor acceptance without
sampling, it is reasonable to think that in all these cases where the optimum sampling plan is
constant, then Acceptance Sampling would be redundant. On the other hand, if it changed dramatically
during time then the use of the switching and/or skip-lot procedures would be essential.
extreme, the evaluation of the standard is adverse and it seems somewhat unfair. Nevertheless,
we choose to examine some combinations of cost elements that lead to such kind of comparisons,
because in practice many times a company should adopt an extreme control policy for a lot,
which is usually not provided by the standard.
Comparing the average cost K(n,c) of the ISO 2859 sampling plans to the optimum K(n*,c*)
The use of the sampling plans recommended by the ISO 2859 instead of the optimum ones,
frequently leads to significant cost burden, which basically varies according to the values of ci, cr,
cd and, consequently, pr. The maximum and the average percentage cost increase max and ,
for p distributed either uniformly or normally between 0.15% and 4% and for every pr investigated,
are presented numerically in Table 3 and graphically in Figure 1. Table 3 includes also the standard
deviation of the percentage cost increase s. It should be noted that every value of max, and
s presented in Table 3 and Figure 1 has been calculated taking into consideration all Ks, for
every N and AQL examined.
[Insert Table 3 about here]
[Insert Figure 1 about here]
The main conclusion is that both max and take their minimum values for a value of pr,
which seems to be close to p , namely 2.075% in the examined distributions. For any other value
of pr, both percentage cost burdens increase. In fact, for values of pr outside the range of p values,
the increase of burdens becomes progressively enormous. The specific conclusion holds for any
other distribution of p that has been examined.
Remark 4: The proper use of the following rules requires estimating as accurately as possible
ci, cr and cd, and calculating pr.
Rule 1: If pr p then trust the recommendations of ISO 2859 more than in any other
case, i.e., the economic loss of using the sampling plans of the ISO 2859 standard is minimum.
Regarding the s values it can be noticed that for the majority of combinations presented in
Table 3, s takes a value close to . This can be explained by the specific form of the frequency
distribution of Ks that have been considered for the determination of both statistical parameters: a
very large frequency of Ks (approaching 50% in some cases) appears in the class close to Kmax,
while another very large frequency (also approaching 50% in some cases) appears in the class
above 0%. It can be easily proved that when a random variable X usually takes two values, for
ab
example a and b, then x
and s
2
n
a x 2 n b x 2
2
2
a x 2 b x 2
2
ab
2
Pa
P ( p )( p )dp is the average probability of accepting a lot when the fraction nonconforming p is
a
reduced compared to the optimum one and thus the percentage loss increases. When the ISO 2859
sampling plan changes, it instantly tends to the optimum one and, thus, peaks of improvement
appear periodically.
[Insert Figure 2 about here]
[Insert Table 5 about here]
On the other hand, when pr is much larger than p (and consequently the optimum inspections
are reduced, i.e., n* 0 and c* n*), the increase of N (for a given AQL) leads to a reduction of
the percentage economic burden of preferring the sampling plans of ISO 2859 instead of the
optimum ones (Figure 3 illustrates this tendency). Note again, that whenever the increase of N
leads to a change of the inspection plan recommended by the standard, an instant increase of the
percentage economic loss takes place. In order to explain this behavior we present in Table 6 similarly to Table 5 - a series of economically optimum inspection plans and the respective series
of plans recommended by ISO 2859. As long as the increase of N does not change the sampling
plan recommended by the standard, the concept of the inspection comes closer to the optimum one,
i.e., reduced inspection or no inspection at all, causing a reduction in the percentage loss. When
the plan recommended by ISO 2859 changes, there is a divergence from the optimum plan and,
thus, peaks of deterioration appear from time to time.
[Insert Figure 3 about here]
[Insert Table 6 about here]
When pr is close to p , the increase of N (for a given AQL) leads to a mixture of the two
behaviors described above. In particular, for small values of N the increase of the lot size leads to a
reduction of the percentage economic loss (Figure 4, region A), while for large values of N the
increase of the lot size leads to an increase of the percentage loss (Figure 4, region B).
[Insert Figure 4 about here]
Rule 2: Assuming that the choice of the AQL value cannot be changed, trust more the
recommendations of ISO 2859 when pr << p , for small values of N, and when pr >> p , for
large values of N.
Additional comments
Studying the impact of changing simultaneously N and AQL on the economic implications of
using the ISO 2859 sampling plans, we have noticed that especially for small values of N, very
small or very large values of AQL are more likely to lead to good economic results (depending on
pr) whereas for large values of N, intermediate values of AQL tend to give sampling plans with
better economic results.
Finally, the impact of a potential differentiation of each cost element on the percentage cost
increase becomes evident, if one keeps in mind that pr increases when ci or cr increases, or cd
decreases and vice-versa.
6.
The use of the sampling plans recommended by the ISO 2859 standard instead of the
economically optimum ones, in order to conduct Acceptance Sampling, results in
significant increase of the quality cost met by companies.
Even in cases where the accurate estimation of the cost elements is not possible, the general
feeling about the economic issues of Acceptance Sampling can lead to the choice of
sampling plans (even through the ISO 2859 standard) that will produce near-optimum
economic results. For instance, it is preferable to choose a tightened inspection plan every
time the cost of rejecting a lot is low and cd is relatively high, while it is better to choose a
reduced sampling plan every time cr cd.
A very important finding that needs to be understood by the practitioners is that although
the sampling process is advisable most of the times, it is not always optimum. For example
it may be optimum not to inspect at all when the quality of the acquired lots is very good
and/or the ci is very high. Similarly, it may be optimum to inspect 100% when ci or cr are
very small, p is high etc. The use of the ISO standard does not allow the practitioner to
adopt either the 100% inspection or the no inspection policy, which are very often
economically optimum.
The application of the three simple rules presented in this paper can lead to a better use of
the ISO 2859 standard.
Bear in mind that the rules for the proper use of the ISO standard, presented in this paper, hold
when the inspection level is the general inspection level II. For other inspection levels (which,
however, are not so popular in practice) similar numerical investigations should be conducted in
order to derive accurate conclusions.
The deeper understanding of the correlation between AQL and pr will contribute to the
improvement of the process of finding (through the ISO 2859 standard) sampling plans that lead to
economically acceptable results. Consequently, the extensive study of this relation constitutes an
interesting topic of future research. Generally, an even more thorough numerical investigation can
reveal information that could help to develop more effective rules of using the ISO 2859 standard.
For example, future research can be directed to the examination of more pdfs of the fraction
nonconforming p, different inspection levels, more combinations of cost elements and/or probably
to the sensitivity analysis of the economic results.
Another interesting area of research would be the study of the complete operation of the ISO
2859 standard, i.e., taking into consideration the procedure of switching to tightened or reduced
inspection and the skip-lot sampling procedures. Moreover, the analysis of this paper has been
based on the assumption that all lots are submitted to rectifying inspection upon rejection. If this is
not the case, then the cost functions should be modified accordingly, offering an appealing area for
fruitful research.
Acknowledgment
The authors would like to thank Mr. Dimitris Hatzinikolaou for his valuable contribution to
this paper. The authors are also grateful to the two reviewers for their valuable remarks that
contributed significantly to the improvement of the paper.
References
1.
Bai, D.S., Riew, M.C. (1984). An Economic Attributes Acceptance Sampling Plan with Three
Decision Criteria, Journal of Quality Technology, 16(3):136-143.
2.
3.
Brown, G.G., Rutemiller, H.C. (1973). A cost analysis of sampling inspection under Military
Standard 105D, Naval Research Logistics Quarterly, 20(1):181-199.
4.
Cassady, R., Nachlas, J.A. (2003). Evaluating and Implementing 3-Level Acceptance
Sampling Plans, Quality Engineering, 15(3):361-369.
5.
Chen, C.-H. (2006). The Modified Pulak and Al-Sultan's Model for Determining the Optimum
Process Parameters, Communications in Statistics - Theory and Methods, 35(10):1767-1778.
6.
Chen, J.W., Chou, W., Wu, H., Zhou, H. (2004a). Designing Acceptance Sampling Schemes
for Life Testing with Mixed Censoring, Naval Research Logistics, 51(4):597-612.
7.
Chen, J.W., Choy, S.T.B., Li K.H. (2004b). Optimal Bayesian Sampling Acceptance Plan with
Random Censoring, European Journal of Operational Research, 155(3):683-694.
8.
ELOT 398.0 (1982). Sampling Procedures and Tables for Inspection by Attributes, Hellenic
Organization for Standardization.
9.
ELOT 398.1 (1982). Sampling Procedures and Tables for Inspection by Attributes Addendum 1, Hellenic Organization for Standardization.
10. Ferrell, W.G., Chhoker, A. (2002). Design of Economically Optimal Acceptance Sampling
Plans with Inspection Error, Computers & Operations Research, 29(10):1283-1300.
11. Gao, Y., Tang, L.C. (2006). Chain Sampling Scheme under Constant Inspection Errors,
Quality And Reliability Engineering International, 22(8):889-903.
12. Gonzlez, C., Palomo, G. (2003). Bayesian Acceptance Sampling Plans Following Economic
Criteria: An Application To Paper Pulp Manufacturing, Journal of Applied Statistics,
30(3):319-333.
13. Hald, A. (1981). Statistical Theory of Sampling by Attributes; Academic Press, London.
14. ISO 2859 (1974). Sampling Procedures and Tables for Inspection by Attributes, International
Organization for Standardization.
15. ISO 2859-1 (1989). Sampling Procedures for Inspection by Attributes - Part 1: Sampling
Plans Indexed by Acceptable Quality Level (AQL) for Lot-by-lot Inspection, International
Organization for Standardization.
16. ISO 2859-1 (1999). Sampling Procedures for Inspection by Attributes - Part 1: Sampling
Schemes Indexed by Acceptance Quality Limit (AQL) for Lot-by-lot Inspection, International
Organization for Standardization.
17. ISO 2859-2 (1985). Sampling Procedures for Inspection by Attributes - Part 2: Sampling
Plans Indexed by Limiting Quality (QL) for Isolated Lot Inspection, International Organization
for Standardization.
18. ISO 2859-3 (2005). Sampling Procedures for Inspection by Attributes - Part 3: Skip-Lot
Sampling Procedures, International Organization for Standardization.
19. ISO 2859-4 (2002). Sampling Procedures for Inspection by Attributes - Part 4: Procedures for
Assessment of Declared Quality Levels, International Organization for Standardization.
20. ISO 2859-5 (2005). Sampling Procedures for Inspection by Attributes - Part 5: System of
Sequential Sampling Plans Indexed by Acceptance Quality Limit (AQL) for Lot-by-lot
Inspection, International Organization for Standardization.
21. ISO 2859-10 (2006). Sampling Procedures for Inspection by Attributes - Part 10: Introduction
to the ISO 2859 series of standards for sampling for inspection by attributes, International
Organization for Standardization.
22. ISO/TR 8550-1 (2007). Guidance on the selection and usage of acceptance sampling systems
for inspection of discrete items in lots - Part 1: Acceptance sampling, International
Organization for Standardization.
23. ISO/TR 8550-2 (2007). Guidance on the selection and usage of acceptance sampling systems
for inspection of discrete items in lots - Part 2: Sampling by attributes, International
Organization for Standardization.
24. Montgomery, D.C. (2001). Introduction to Statistical Quality Control; John Wiley, New York.
25. Nikolaidis, Y., Nenes, G. (2005). Economic Evaluation of Sampling Schemes that are
Proposed by Acceptance Sampling Standards, In Standardization in an Enlarged Europe,
Proceedings of the 10th EURAS Workshop, Tallinn, Estonia, June 2-3, Coenen, H., Grhndal,
J., Jakobs, K., Valdlo, T., Eds.; 111-116.
26. Selinger, K.A. (1995). Inspection by variables as an acceptance criterion in bioanalysis a
proposal, Journal of Pharmaceutical and Biomedical Analysis, 13(12):1427-1436.
27. Stout, Q.F., Hardwick, J. (2005). Optimal Screening Designs With Flexible Cost And
Constraint Structures, Journal of Statistical Planning and Inference, 132(1-2):149-162.
28. Tagaras, G. (2001). Statistical Quality Control (in Greek); Zitis, Thessaloniki.
29. Tagaras, G., Lee, H.L. (1987). Optimal Bayesian Single-Sampling Attribute Plans with
Modified Beta Prior Distribution, Naval Research Logistics, 34(6):789-801.
30. Taylor, W.A. Acceptance sampling update, Appeared in MDDI (Medical Device and
Diagnostic Industry), October 1995, Canon Communications, 17(10):92-108.
31. Wall,
M.S., Elshennawy,
A.K.
(1989).
Table 1: Combinations of cost elements that were examined in the numerical investigation
Combination
Cost elements
pr (%)
ci = 1, cr = 100, cd = 1100
0.10
ci = 1, cr = 50, cd = 450
0.25
ci = 1, cr = 50, cd = 300
0.40
ci = 1, cr = 100, cd = 243
0.70
ci = 1, cr = 100, cd = 180
1.25
ci = 1, cr = 80, cd = 135.5
1.80
ci = 1, cr = 80, cd = 122.5
2.35
ci = 1, cr = 80, cd = 114.5
2.90
ci = 1, cr = 50, cd = 79
3.45
10
ci = 1, cr = 50, cd = 70
5.00
11
ci = 1, cr = 10, cd = 20
10.00
Table 2: Illustrative results for combination 7 of cost elements, selected values of AQL and p ~
U(0.0015, 0.04)
AQL
0.40%
Optimum values
n*
c*
K(n*,c*)
7.63
12.71
17.79
10
12
14
17
20
1.50%
K(n,c)
(%)
7,98
4.65
13,30
4.65
18,62
4.65
25.42
26,60
30.50
31,92
35.59
43.21
50.84
24
61.01
27
38
49
55
70
6.50%
K(n,c)
(%)
K(n,c)
(%)
7,98
4.65
13,30
4.65
7.86
4.65
12.93
18,62
4.65
4.65
18.00
4.65
4.65
26,33
4.65
31,37
4.65
25.61
3.57
4.65
30.68
37,24
4.65
2.86
36,42
4.65
35.76
2.34
45,22
4.65
53,20
4.65
43,99
4.65
44.12
1.80
51,56
4.65
51.74
63,84
4.65
1.42
61,65
4.65
61.89
1.07
68.63
71,82
96.59
32
100,15
4.65
69,23
4.65
69.51
0.87
4.65
96,98
3.68
97.43
124.53
32
0.41
127,70
4.66
124,74
2.55
125.35
0.17
139.74
32
177.70
32
142,73
4.69
139,88
2.14
13
141.28
0.10
180,30
4.78
177,73
1.46
13
179.39
85
215.58
0.02
32
217,87
4.88
215,58
1.06
13
217.50
0.00
100
10
120
12
253.41
32
255,45
4.97
32
255,60
0.81
20
256.50
0.87
303.77
32
305,54
5.08
32
305,75
0.58
20
307.32
140
14
0.65
354.08
32
355,64
5.17
32
355,89
0.44
20
358.14
0.51
160
215
16
404.35
32
405,74
5.26
32
406,03
0.34
32
410.46
0.42
19
542.45
32
543,50
5.24
32
543,92
0.19
32
550.26
0.27
270
52
680.13
32
681,26
4.88
32
681,81
0.17
32
690.06
0.25
300
55
754.65
32
756,41
4.82
50
760,28
0.23
50
768.46
0.75
390
61
977.93
32
981,84
4.73
50
986,10
0.40
50
997.22
0.84
480
65
1200.95
32
1207,27
4.70
50
1211,92
0.53
50
1225.98
0.91
550
68
1374.31
125
1387,72
4.69
80
1388,13
0.98
80
10
1407.47
1.01
850
114
2112.83
125
2132,59
4.88
80
2138,34
0.94
80
10
2170.03
1.21
1150
160
2848.51
125
2877,45
5.07
80
2888,55
1.02
80
10
2932.59
1.41
1400
164
3460.25
125
3498,17
5.21
125
3524,19
1.10
125
14
3573.38
1.85
2200
254
5410.19
125
5484,47
5.60
125
5526,82
1.37
125
14
5606.88
2.16
3000
301
7354.29
125
7470,78
5.86
125
7529,45
1.58
125
14
7640.37
2.38
3500
345
8567.16
200
8687,35
6.00
200
8733,36
1.40
200
21
8920.19
1.94
6600
521
11
16067.11
200
16348,43
6.50
200
16437,67
1.75
200
21
16800.00
2.31
9700
651
14
23547.89
200
24009,52
6.77
200
24141,99
1.96
200
21
24679.81
2.52
12500
738
16
30295.77
315
31044,73
6.93
315
10
30877,06
2.47
200
21
31797.05
1.92
22500
996
22
54359.44
315
55834,91
7.25
315
10
55529,64
2.71
200
21
57215.79
2.15
32500
1209
27
78392.95
315
80625,09
7.42
315
10
80182,20
2.85
200
21
82634.53
2.28
40000
1377
31
96406.91
500
99243,52
9.27
500
14
97561,35
2.94
200
21
101698.60
1.20
92500
2101
48
222378.40
500
229381,70
9.53
500
14
225463,80
3.15
200
21
235147.00
1.39
145000
2609
60
348248.30
500
359519,90
9.63
500
14
353366,20
3.24
200
21
368595.40
1.47
160000
2740
63
384201.20
800
400154,20
9.65
800
21
387015,80
4.15
200
21
406723.5
0.73
Table 3: max, and s depending on pr, for p ~ U(0.0015, 0.04) and p ~ TN(0.0015, 0.04)
Combination
of cost
U(0.0015, 0.04)
TN(0.0015, 0.04)
pr (%)
elements
max (%)
(%)
s (%)
max (%)
(%)
s (%)
0.10
641.47
229.98
222.78
636.57
234.67
220.57
0.25
358.68
126.87
124.61
355.27
130.08
123.37
0.40
207.96
72.07
71.87
204.46
74.11
71.16
0.70
66.71
22.09
22.71
64.43
22.72
22.10
1.25
26.34
7.64
8.24
23.67
7.49
7.64
1.80
13.77
3.30
3.61
10.67
2.52
2.71
2.35
9.65
2.95
2.12
7.07
2.53
1.78
2.90
12.89
5.92
4.31
11.96
6.43
4.26
3.45
24.30
12.59
8.56
24.30
13.73
8.48
10
5.00
40.28
22.43
13.87
40.28
23.64
13.90
11
10.00
190.96
110.49
65.41
190.96
114.39
65.55
(n*,c*)
Relevant
change of (n,c)
figure
pr << p
Tightened
(instantly)
pr >> p
Reduced
(instantly)
Neither
pr p
tightened nor
reduced
Table 5: Economically optimum inspection plans (n*,c*) for p r 0.1% and respective plans
recommended by ISO 2859 (n,c) for AQL = 0.4% - p ~ U(0.0015, 0.04)
n*
c*
Pa (%)
Pa (%)
300
300
5.49
32
54.44
390
390
3.69
32
54.44
480
480
2.62
32
54.44
550
550
2.06
125
36.48
850
850
0.85
125
36.48
1150
1150
0.40
125
36.48
1400
1400
0.23
125
36.48
2200
2200
0.04
125
36.48
3000
2991
0.01
125
36.48
3500
3467
0.00
200
34.68
6600
5407
0.00
200
34.68
9700
6094
0.00
200
34.68
Table 6: Economically optimum inspection plans (n*,c*) for p r 5% and respective plans
recommended by ISO 2859 (n,c) for AQL = 0.4% - p ~ U(0.0015, 0.04)
n*
c*
Pa (%)
Pa (%)
300
100.00
32
54.44
390
100.00
32
54.44
480
100.00
32
54.44
550
100.00
125
36.48
850
100.00
125
36.48
1150
100.00
125
36.48
1400
100.00
125
36.48
2200
100.00
125
36.48
3000
100.00
125
36.48
3500
100.00
200
34.68
6600
100.00
200
34.68
9700
100.00
200
34.68
Relevant
change of c from 0 to 1
figure
(instantly)
(instantly)
(n*,c*)
Increase of AQL
pr << p
Tightened
pr >> p
Reduced
Neither
pr p
tightened nor
reduced
Table 8: Rule 3
Dont use the standard! Just reject the lot.
pr < pmin
Approximate areas of pr
Large values of N:
Choose AQL pr
pr p
Choose AQL pr
No simple rule
Dont use the standard! Just accept the lot.
pmax < pr
Alternatively choose the maximum value of AQL
Figure 1: max and depending on pr, for p ~ U(0.0015, 0.04) and p ~ TN(0.0015, 0.04)
700%
p ~ U(0.0015, 0.04) and
p ~ TN(0.0015, 0.04)
600%
500%
Maximum
Average
400%
300%
200%
100%
0%
0%
2%
4%
6%
8%
10%
pr
800%
p ~ U(0.01, 0.07)
700%
600%
400%
300%
200%
100%
Lot size
160000
40000
12500
3500
1400
550
300
160
100
55
27
17
10
0%
3
500%
250%
35%
30%
150%
p ~ U(0.0005, 0.005)
20%
15%
100%
10%
50%
5%
p ~ U(0.0015, 0.04) & p ~ TN(0.0015, 0.04)
160000
40000
3500
1400
12500
Lot size
550
300
160
100
55
27
17
10
0%
0%
25%
-p ~ U(0.0005, 0.005)
200%
Figure 4: depending on N, for p r 2.35% , p ~ U(0.0015, 0.04) and various values of AQL
6,00%
5,00%
B
AQL = 0.4%
3,00%
2,00%
1,00%
Lot size
2E+05
40000
12500
3500
1400
550
300
160
100
55
27
17
AQL = 6.5%
10
0,00%
4,00%
Figure 5: depending on AQL, for selected values of N, p r 0.1% and p ~ U(0.0015, 0.04)
700%
N = 160,000
600%
500%
400%
300%
N =3
200%
100%
AQL
6.5%
4%
2.5%
1.5%
1%
0.65%
0.4%
0.25%
0.15%
0.1%
0.065%
0%
45%
40%
N =3
35%
25%
20%
15%
10%
5%
N = 160,000
AQL
6.5%
4%
2.5%
1.5%
1%
0.65%
0.4%
0.25%
0.15%
0.1%
0%
0.065%
30%