Académique Documents
Professionnel Documents
Culture Documents
1
INTRODUCTION
ing as compared with 100 percent inspection. However, sampling has additional adva
f a sample gives information then and there about the quality of the pieces in th
known process now can tell on the product. We already know about the inspected pro
bors of the inspected pieces; acceptance sampling is also valid because the uninsp
not. Once the series of samples tells on the process by certifying stability, we
eptable limits
defective. This refers to the acceptability of units of product for a wide range o
o of defects per unit. This refers to the number of defects found in the units ins
eviation.
fit for use and product which does not meet specification requirements. These ar
nts are then summarized into a simple statistic (e.g., sample mean) and the observe
rotection. The Table shows a comparison of variables sample sizes necessary to ach
ria for accepting the lot,” and a sampling scheme as “a combination of sampling pl
may be chosen.”
umer, it is usually best to use the published procedures and standards. This avoid
od” lots and of accepting “bad” lots, with associated serious consequences, produce
lity level which for purposes of sampling inspection is the limit of satisfactory
rejected by the sampling plan. A consumer’s risk of 0.10 is common and LTPD has be
ling plan will accept the lot. The Figure next slide shows an ideal OC curve for
and to accept the lot if no more than four defective pieces were found. Such curve
p = 3.0 defectives in the sample. The table shows for r = 4 and np = 3.0, the pro
units inspected per lot in sampling inspection, ignoring any 100 percent inspecti
on the initial sample, and consequently being forced to inspect a second, third, e
imply:
ASN = n1 + (1 - Pa1 ) n2
than pb, the total cost will be lowest with sampling inspection or no inspection. I
tive will be greater than 5 percent, then 100 percent inspection should be used.
lity level
judging conformance.
used for judging the conformance of a single unit of product to standard are some
he sample. Since defects differ greatly in seriousness, the sampling plans must so
lowed.
rming lots are subsequently reviewed to judge fitness for use, the review board g
ame product. Nine of these produce perfect product. The tenth machine produces 100
fact that lot formation so strongly influences outgoing quality and inspection e
e of the lot. In those cases where the inspector has knowledge of how the lot was
units of product has an equal chance of being the next unit selected for the samp
and then proceeding in some chosen direction (up, down, right, left, etc.) to obtain
h are detrimental to good decision making. The more usual biases consist of:
hance of being accepted; i.e., the operating-characteristic curves can be made quit
lot is reached on the basis of this first sample if the number of nonconforming u
to accept or reject the lot. The chief advantage of multiple sampling plans is a
Such schemes employ 100 percent inspection (screening), with nonconforming items r
oduction based on the AOQL concept. These plans generally start with 100 percent
, at the rate of 1 unit in 10, is then instituted. If a defective is found, 100 per
the start until i successive units are found free of defects. When sampling is in
specified number of defectives is found during any one sequence of 100 percent i
ip-lot plans are based are much like those for chain-sampling Plans.
ampling) are:
osen at random for inspection and inspected to the same n and c. As long as all lo
t lot under inspection can also be accepted if one defective unit is observed in
sadvantage is that moderate changes in quality are not easily detected. However, m
onal procedure for defining lots. A common procedure was to accumulate product un
ejected according to the cumulative sum of defectives observed. The Figure on next
s a second parameter of the scheme. When Sm < 0, return the accumulation to zero.
eject product and restart the accumulation at h + h' (h‘ is a third parameter of t
h. When Sm h + h', return calculation to h + h' and continue rejecting product un
ccept product and restart accumulation at zero.
ss than or equal to that AQL. These plans are devised for producer’s protection or
a sensible grouping and indexing of plans for use.
plans are chosen such that the consumer’s risk of accepting a submitted lot of p
tion that no defectives in the sample implies there are no defectives in the rest
he-art quality levels are indeed not zero, plans with nonzero acceptance numbers a
observed value is compared with an allowable value defined in the plan. A decisio
s for percent nonconforming may be used. These plans relate the proportion of indi
iation σ known. Suppose the following sampling plan is used to test against an u
indicated by . Means of samples of size n are, then, distributed about , as shown; so the pr
limit is set inside the specification limit. Samples are selected and gagged aga
ze can, however, be made to apply to several lots jointly if the lots can be shown
control chart analysis for uniformity in level and variation before they are com
laboratory testing.
tween packages, and if the quality of each package in the sample is measured, then
n part at least, to form one or more composite samples that are tested separately.
packages.
erial may be used. It is conceivable that interest in bulk material may also at t
posite sample. It may take the form of hand-quartering or riffling or the like.
eristic.
on the test-unit. Any single measurement is a random sample of one from this infi
is designated the mean of the lot. It is postulated that this mean will be the sam
define the mean of a lot. It is assumed that the segment is so large relative to t
sis here is on the feedback of information necessary for the proper application,
en used in a system of
his is much as the old time inspector varied the inspection depending on quality