ACCEPTANCE SAMPLING
In the preceding material we have looked
at various
techniques
and their application for controlling the quality of the product as
it is
being produced. Our primary concern was
to us
e the information to help make decis
ions
as
to what action s
hould be taken in continuing or correcting the proces
s
as
it was
running. We s
hall now look at the problem of making decis
ions
about material s
ubmitted for our acceptance. This
may be material from an outs
ide s
upplier or from a preceding fabrication or proces
s
ing department for which certified control charts
relating to the proces
s
are not available. TECHNIQUES USED FOR ACCEPTING
MATERIAL There are s
everal methods
which have been us
ed at various
times
, and for various
reas
ons
, for the acceptance of material. Among thos
e which we s
hall dis
cus
s
are:
1. No tes
ting or ins
pection at all 2. 100% s
orting (detailing)
3. Spot-checking 4. Cons
tant percentage s
ampling 5. Scientific s
ampling As
you would expect, the mos
t practical in the long run involves
the us
e of s
cientific s
tatis
tical s
ampling plans
. Occas
ionally materials
will be received and s
ent directly to s
torage, proces
s
ing or as
s
embly without being checked. Very little or nothing is
known about the quality of s
uch material until it is
needed for us
e. This
is
not a very oppertune time to dis
cover nonconforming material. At the other end of the s
pectrum is
the other extreme involving 100% ins
pection. This
is
s
ometimes
needed for critical parts
where cons
iderable ris
k is
involved. On the other hand, if the ris
k is
not great, 100% ins
pection can be quite cos
tly. Further, 100% ins
pection does
not as
s
ure that all product conforms
. Experience s
hows
that 100% ins
pection is
only about 85 to 90% efficient due to fatigue and human error inherent in s
uch repetitive activities
. To demons
trate the above fact let's
do a little experiment. Read the following paragraph ONLY ONCE and circle each "F" that you s
ee. Count the circles
without rereading the paragraph. The people that live in the United States
enjoy the privilege of having a plentiful s
upply of food. Many of us
eat s
o much food that we gain weight exces
s
ively. It is
folly for us
to eat s
o much that we get fat while people in other countries
are s
tarving. Perhaps
our abundance of food is
due to well planned s
oil cons
ervation. Regardles
s
of how we obtain our s
upply of food, the average American hus
band likes
a wife who is
able to maintain a flattering figure for hers
elf. If you found all of the F's
the firs
t time, this
is
the equivalent to 100% ins
pection. If it was
neces
s
ary to read the paragraph the s
econd time, this
is
equivalent to 200% ins
pection. There are nineteen F's
in the paragraph and if you have to read it again to find them all that means
that you have performed 300% ins
pection. Who's
going to pay for all that ins
pecting?? In an attempt to arrive at a compromis
e between no checking and 100% checking, s
ome people have res
orted to s
pot-checking random s
hipments
. This
will s
top s
ome defective material but it is
only partially effective s
ince many lots
will be accepted without a check. Pos
s
ibly the mos
t frequently us
ed acceptance technique has
been the ins
pection of an arbitrary cons
tant-percentage s
ample from each s
ubmitted lot of material. The mos
t notorious
s
uch plan has
been the 10% s
ample. Regardles
s
of the s
ize of the lot, a s
ample of 10% is
tes
ted and the res
ults
of that tes
t is
us
ed to determine the quality of the entire lot. This
technique is
s
omewhat better than the preceding plans
, but it s
till has
certain weaknes
s
es
. For s
ome material, 10% is
too s
mall a s
ample, while for others
it may be to large. Such plans
place a burden of over ins
pecting on large lot s
izes
. As
we s
hall s
ee later, a 10% s
ample from 1000 pieces
is
far more s
tringent than is
a 10% s
ample of 50 pieces
. In any s
ampling technique, certain ris
ks
of making a wrong decis
ion is
inherent in the s
ys
tem. Defective material may be accepted, or acceptable material may be rejected. All too often thos
e us
ing arbitrary percentage s
ampling plans
as
s
ign a rejection number of one nonconformity with little or no idea of the ris
ks
inherent in the plan. Scientific s
ampling plans
pos
s
es
s
many of the s
ame characteris
tics
as
the arbitrary plans
in providing a compromis
e between no ins
pection and 100% ins
pection. They have the advantage of providing known ris
ks
of erronious
decis
ions
and they do not dis
criminate agains
t large lot s
izes
. PRELIMINARY CONSIDERATIONS Let's
go back to our paragraph with the F's
for a minute. If we were to cons
ider this
to be a lot of a certain product with the F's
repres
enting nonconforming material we find that there are 19 nonconforming items
among the 383 items
in the lot. This
means
that the lot is
nearly 5% (4.96)
nonconforming. From tables
of s
ampling plans
, which we will explain later, we find that if we could tolerate product which contained no more than 1% nonconforming parts
, we s
hould take a s
ample of 50 items
and accept if we find no more than one nonconforming part in the s
ample. If we count the firs
t 50 letters
in our product lot we find that there are no F's
s
o we accept the entire lot. But we know that it is
nearly 5% nonconforming. This
brings
up another element which is
extremely important in s
ampling. The s
ample mus
t be repres
entative of the total lot, therefore, we mus
t take an unbias
ed random s
ample. This
means
that any item in the lot mus
t have an equal opportunity of being s
elected in the s
ample. If we thoroughly mix the letters
and then draw 50 from the mixture we might obtain a s
ample s
imilar to the one s
hown below. The s
ample contains
4 nonconformities
s
o we would reject the lot at the 1% level. However, the plan for a 6% lot calls
for a s
ample of 50 and permits
us
to accept if we find 6 or les
s
nonconformities
in our s
ample. As
can be s
een we could accept the lot if we feel that we can work with material that is
no wors
e than 6% nonconforming. G
TOTU IONFE SABED AWLAP FHSAK
OG
ATH IFALP YWEPA YTMLF OTITA It s
hould be obvious
that before any s
ampling plan can be us
ed the characteris
tics
to be checked mus
t be clearly identified and adequate ins
truments
provided to check thos
e characteris
tics
. Care mus
t be taken to provide an unbias
ed random s
ample of the s
ubmitted material. Unles
s
the data is
accurate, hones
t and random, the res
ults
will have no merit or advantage over the decis
ion to take no s
amples
. A s
tatis
tical s
ampling plan will not enhance s
loppy or erroneous
data. THE OPERATING
CHARACTERISTIC CURVE One of the mos
t valuable aids
in s
electing a s
ampling plan is
the OPERATING
CHARACTERISTIC CURVE, commonly referred to as
the O. C. Curve. This
is
a plotted curve which s
hows
the probability (ris
k)
of accepting product at varying quality levels
under a given plan. Each plan has
its
own O. C. Curve. Shown below is
an O. C. Curve for a s
ample of 10 items
taken from a lot of 100 items
. If the s
ample contains
0 or 1 nonconformities
, the lot is
accepted, if it has
two or more the lot is
rejected. The s
ymbolic form in which this
plan is
written is
as
follows
:
N = 100 n = 10 c = 1 where is
the lot s
ize, n is
the s
ample s
ize and c is
the acceptance number. (INSERT O.C.CURVE FOR THIS PLAN)
Since the O. C. Curve s
hows
the probability of accepting a lot at a given percent nonconforming, we can s
ee that if the lot is
perfect, we s
hould expect all lots
to be accepted. However if the lots
contain 10% nonconforming product we would expect to accept only about 74% of the s
ubmitted lots
. This
means
that we reject about one fourth of the lots
on ins
pection. It can be s
een that this
threat of rejection of poor quality product can have a powerful influence on the s
upplier. ACCEPTABLE QUALITY LEVEL -- AQL Anytime the decis
ion is
made to us
e an acceptance procedure it becomes
neces
s
ary to es
tablis
h jus
t what quality level is
cons
idered to be acceptable. It mus
t be realized that it is
inevitable that there is
a degree of variability in any proces
s
and that s
hort of extremely redundant ins
pection we will have s
ome percentage of nonconforming product in the lots
. Although we would prefer 100% perfect product we might not be able to pay for it when we get it. There is
no s
imple rule for es
tablis
hing an AQL. One aid is
to examine material accepted and us
ed in the pas
t and adjus
t this
to a value which can be lived with. Once we have es
tablis
hed this
value we have a meas
ure of what cons
titutes
"good quality." If our s
ampling plan operates
s
atis
factorily, we want to accept nearly all of the lots
which are as
good or better than our AQL and reject nearly all of thos
e which are wors
e than our acceptable level. Notice that we want to accept NEARLY all of the good lots
. This
is
in recognition of the ris
ks
involved in s
ampling. Even though we s
hould des
ire all lots
this
good to be accepted, we know that s
ampling fluctuations
will occas
ionally caus
e us
to make an erroneous
decis
ion. To be more s
pecific, let's
s
ay that we want our s
ampling plan to accept at leas
t 95% of the lots
which are as
good as
, or better than, the AQL. We have thus
acknowledged a ris
k of 5% of making an erroneous
decis
ion at a particular incoming quality level and have s
pecified one point on our O. C. Curve. Since the ris
k we have s
pecified is
the ris
k of rejecting material which we have clas
s
ified as
acceptable, we call this
the "PRODUCER'S RISK
". The diagram on the next page is
prepared for an AQL of 3% and a Producer's
ris
k of 5%. We refer to this
as
the producers
ris
k becaus
e the producer is
the one who is
hurt if the lot is
in fact as
good as
or better than the AQL and is
rejected. (INSERT G
RAPH OF O. C. CURVE SHOWING
5% PRODUCER'S RISK
AT AQL =3%)
REJECTABLE QUALITY LEVEL -- RQL By s
electing the AQL, we have s
pecified a meas
ure of acceptable quality. n a s
imilar manner, we mus
t determine what quality is
s
o bad that if we were to receive it we might expect to encounter great difficulty in us
ing it. This
is
a meas
ure of unacceptable quality and is
commonly referred to as
the Rejectable Quality Level, or RQL. Sometimes
this
is
referred to as
the Lot Tolerance Percent Defective. LTPD Again we mus
t recognize the ris
ks
inherent in s
ampling plans
. We know that s
ampling fluctuations
will frequently caus
e us
to accept material which we don't want, but we can limit this
ris
k to s
ome s
pecified amount. Let's
s
uppos
e that we wouldn't want to accept more than 10% of the lots
pres
ented if they were as
bad as
, or wors
e than, the . Since we are concerned with the ris
k of accepting material which is
unacceptable, we call this
ris
k the Cons
umer's
Ris
k. By s
pecifying the RQL and Cons
umer's
Ris
k, we have located another point on the O. C. Curve. The following diagram has
been prepared for an RQL of 8% and a Cons
umer's
Ris
k of 10%. (INSERT O.C.G
RAPH SHOWING
10% CONSUMER'S RISK
)
It s
hould be noted that it is
common to s
et the Producer's
ris
k
at 5% and the Cons
umer's
ris
k at 10% and then to adjus
t the AQL and RQL to obtain the protection we des
ire. A s
ampling plan which would meet the criteria we've placed upon it by our s
election of an AQL and RQL would require a s
ingle s
ampling plan with a s
ample s
ize in exces
s
of 200. One plan which would s
atis
fy our requirements
would be a s
ample s
ize of 225 and an acceptance number of 11. This
means
that we would have to examine 225 pieces
and would then accept the lot if the s
ample contained 11 or les
s
nonconforming pieces
and reject it if it contained 12 or more. This
plan is
illus
trated by the following O. C. Curve. (INSERT 0. C. CURVE FOR N=225 AND C=11)
THE FALLACY OF A CONSTANT PERCENTAG
E SAMPLE There is
an error made by thos
e who believe that a cons
tant percentage s
ample s
ubjects
material to a fair tes
t regardles
s
of the s
ize of the lot. A cons
tant percentage s
ample inflicts
a greater penalty on large lots
than it does
on s
mall lots
. This
fact is
demons
trated in the diagram on the following page. The fact is
that a cons
tant s
ample s
ize will provide a more cons
is
tent degree of protection than will the cons
tant percentage s
ample.