Knowing What to Do

Knowing How to Do It
Getting Better Every Day

Acceptance Sampling Webinar

20101129

Acceptance Sampling
I

Acceptance Sampling Webinar

20101129

What you will learn

The purpose of Sampling
How to draw a statistically valid Sample
How to Develop a Sampling Plan
How to construct an O-C curve for your sampling

plan
How to use (and understand) ANSI/ASQ Z1.4
How to use ANSI/ASQ Z1.9
Assessing Inspection Economics
Acceptance Sampling Webinar
20101129

What is Sampling
Sampling refers to the practice of evaluating
(inspecting) a portion -the sample - of a lot the
population for the purpose of inferring information
about the lot.
Statistically speaking, the properties of the sample
distribution are used to infer the properties of the
population (lot) distribution.
An accept/reject decision is normally made based on
the results of the sample
Sampling is an Audit practice
Acceptance Sampling Webinar
20101129

Why Sample?
Economy
Less inspection labor
Less time
Less handling damage
Provides check on process control
Fewer errors ???
i.e. inspection accuracy

Acceptance Sampling Webinar

20101129

What does Sampling not do?

Does not provide detailed information of lot quality
Does not provide judgment of fitness for use (of

rejected items)
Does not guarantee elimination of defectives any
AQL permits defectives

Acceptance Sampling Webinar

20101129

Sampling Caveats
Size of sample is more important than percentage of lot
Only random samples are statistically valid
Access to samples does not guarantee randomness
Acceptance sampling can place focus on wrong place
Supplier should provide evidence of quality
Focus should be on process control

Misuse of sampling plans can be costly and misleading.

No such thing as a single representative sample

Acceptance Sampling Webinar

20101129

Representative Sample?
There is no such thing as a single
representative sample
Why?
Draw repeated samples of 5 from a normally

distributed population.
Record the X-bar (mean) and s (std.dev) for each
sample
What is the result?
Acceptance Sampling Webinar
20101129

Distribution of Means

The Distribution of Means obeys normal distribution regardless of

distribution of parent population.
Acceptance Sampling Webinar
20101129

Standard Error of the Mean

Central Limit Theorem

The relationship of the standard deviation of sample

means to the standard deviation of the population
Note: For a uniform distribution, Underestimates error by 25% with
n=2, but only by 5% with n=6

Acceptance Sampling Webinar

20101129

10

The Random Sample

At any one time, each of the remaining items in the
population has an equal chance of being the next
item selected
One method is to use a table of Random Numbers
(handout from Grant & Leavenworth)
Enter the table Randomly ( like pin-the-tail-on-thedonkey)
Proceed in a predetermined direction up, down, across
Discard numbers which cannot be applied to the sample
Acceptance Sampling Webinar
20101129

11

Random Number Table

Acceptance Sampling Webinar

20101129

Source: Statistical Quality Control by Grant &

Leavenworth
12

Stratified Sampling
Random samples are selected from a homogeneous lot.

Often, the parts may not be homogeneous because they were

produced on different machines, by different operators, in
different plants, etc.
With stratified sampling, random samples are drawn from
each group of processes that are different from other
groups.

Acceptance Sampling Webinar

20101129

13

Selecting the Sample

Wrong way to select sample
Judgement: often leads to Bias
Convenience
Right ways to select sample
Randomly
Systematically: e.g. every nth unit; risk of bias occurs
when selection routine matches a process pattern

Acceptance Sampling Webinar

20101129

14

The O-C Curve

Operating Characteristic Curve

Ideal O-C Curve

Pa

Percent Defective
Acceptance Sampling Webinar
20101129

15

The Typical O-C Curve

Acceptance Sampling Webinar

20101129

16

Sampling Terms
AQL Acceptable Quality Level: The worst quality

level that can be considered acceptable.

Acceptance Number: the largest number of defective
units permitted in the sample to accept a lot usually
designated as Ac or c
AOQ Average Outgoing Quality: The expected

quality of outgoing product, after sampling, for a

given value of percent defective in the incoming
product. AOQ = p * Pa
Acceptance Sampling Webinar
20101129

17

Sampling Terms (cont.)

AOQL Average Outgoing Quality Level: For a

given O-C curve, the maximum value of AOQ.

Rejection Number smallest number of defective
units in the sample which will cause the lot to be
rejected usually designated as Re
Sample Size number of items in sample usually

designated by n
Lot Size number of items in the lot (population)
usually designated by N
Acceptance Sampling Webinar
20101129

18

Sampling Risks
Producers Risk : calling the population bad

when it is good; also called Type I error

Consumers Risk : calling the population good

when it is bad; also called Type II error

Acceptance Sampling Webinar

20101129

19

Sampling Risks (cont)

Acceptance Sampling Webinar

20101129

20

Acceptance Sampling
II

Acceptance Sampling Webinar

20101129

21

Constructing the O-C curve

We will do the following O-C curves
Use Hyper-geometric and Poisson for each of the
following
N=60, n=6, Ac = 2
N=200, n=20, Ac = 2
N=1000, n=100, Ac = 2
N=1000, n=6, Ac = 2

Lets do k (Ac, c - # of successes) = 0 first

Acceptance Sampling Webinar

20101129

22

Hyper-geometric

The number of distinct combination of n items

taken r at a time is

Acceptance Sampling Webinar

20101129

23

Hyper-geometric (cont)
= (DCk

Cn-k) / NCn

Nq

Construct the following Table

p
D=Np P(k=0) P(k=1) P(k=2) P(k 2)
0%
1%
2%
3%
etc.
A Hyper-geometric calculator can be found at www.stattrek.com

Note: The Hyper-geometric distribution applies when the population, N, is small

compared to the sample size, however, it can always be used. Sampling is done
without replacement.
Acceptance Sampling Webinar
20101129

24

Hypergeometric Calculator

Acceptance Sampling Webinar

20101129

25

Hypergeometric Calculator
Example: p=0.02, k=0, N=100,
n=10

Acceptance Sampling Webinar

20101129

26

Hypergeometric Calculator
Example: p=0.02, k=0, N=100, n=10

Acceptance Sampling Webinar

20101129

27

Hypergeometric Calculator
Example: p=0.02, k=0, N=100, n=10
P (k=0) = 0.809091
P (k=1) = 0.181818
P (k=2) = 0.009091
----------------------P(k2) = 1.0

Acceptance Sampling Webinar

20101129

28

Acceptance Sampling Webinar

20101129

29

Acceptance Sampling Webinar

20101129

From QCI-CQE Primer 2005, pVI-9

30

Poisson
Construct the following Table, using the Poisson Cumulative Table
p
np
P (k 2)
0%
1%
2%
3%
4%
etc.

Compare. When is Poisson a good approximation

Use the Poisson when n/N0.1 and np 5.
Acceptance Sampling Webinar
20101129

31

Poisson Calculator
Example: p=0.02, n=10, c=0

X=k, the number of successes in the sample, i.e. c

Acceptance Sampling Webinar

20101129

32

Poisson Calculator
Example: p=0.02, n=10, c=0

Mean = np

Acceptance Sampling Webinar

20101129

33

Poisson Calculator
Example: p=0.02, n=10, c=0

TRUE for cumulative, i.e. k; FALSE for probability mass function, i.e.p(x=k)

Acceptance Sampling Webinar

20101129

34

From QCI-CQE Primer 2005, pVI-8

Acceptance Sampling Webinar
20101129

35

From QCI-CQE Primer 2005, pVI-8

Acceptance Sampling Webinar
20101129

36

Acceptance Sampling Webinar

20101129

From QCI-CQE Primer 2005, pVI-9

37

O-C Curve & AOQ

Determine the O-C curve.
Prepare the following Table using the Poisson distribution
p
0%

Pa

AOQ = p * Pa

1%
2%
3%
etc
Graph the results: Pa and AOQ vs p.
Acceptance Sampling Webinar
20101129

38

OC Curve & AOQ (2)

Acceptance Sampling Webinar

20101129

39

OC Curve & AOQ (3)

Acceptance Sampling Webinar

20101129

40

Acceptance Sampling
III

Acceptance Sampling Webinar

20101129

41

Questions
1. What if this AOQ is not adequate?
2. What if you would like to add a 2nd sample when the

first sample fails?

Example
OC curve after 1st Sample:
p=0.02, n=30, N=500, c (Ac)=0, Re=2

OC curve after 2nd Sample (of 30 more):

p=0.02, n=60, N=500, c (Ac)= 1, Re=2

Acceptance Sampling Webinar

20101129

42

Hypergeometric Multiple Sampling

D=Np

N=

500

500

500

500

n=

30

60

60

60

Nq=N-Np

P(k=0)

P(k=0)

P(k=1)

0
1

P(k 1)
1

0.00

500

0.01

495

0.73

0.53

0.36

0.89

0.02

10

490

0.54

0.28

0.38

0.66

0.03

15

485

0.39

0.14

0.30

0.44

0.04

20

480

0.28

0.07

0.21

0.28

0.05

25

475

0.20

0.04

0.14

0.17

0.06

30

470

0.15

0.02

0.08

0.10

0.07

35

465

0.11

0.01

0.05

0.06

Acceptance Sampling Webinar

20101129

43

Hypergeometric Multiple Sampling

Acceptance Sampling Webinar

20101129

44

ANSI/ASQC Z1.4-1993
Mil-Std 105

Sampling for Attributes; 95 page Document

Pas from 83% to 99%
Information necessary: N, AQL, Inspection Level
How to Use
Code Letters
Single, Double, Multiple Plans
Switching Rules
Obtain: n, Ac, Re,
O-C Curves
Acceptance Sampling Webinar
20101129

45

ANSI/ASQC Z1.4-1993
Exercises

N=475, AQL = 0.1%, Single Plan, Normal

What is Code Letter
What is Sample Size,
What is Ac, Re
Repeat for Tightened Inspection
Repeat for Reduced Inspection

Note: 0.1% is 1000 ppm

Acceptance Sampling Webinar
20101129

46

Z1.4 Code Letters

I-Reduced, II-Normal, III-tightened |||| For N=475, Normal, code letter is H

Acceptance Sampling Webinar

20101129

47

Z1.4 Single Plan Normal Insp.

Table II-A

n=125,

New code Letter K

Acceptance Sampling Webinar

20101129

48

Z1.4 O-C Curve for Code Letter K

Table X-K

Acceptance Sampling Webinar

20101129

49

Z1.4 Switching Rules

Acceptance Sampling Webinar

20101129

50

ANSI/ASQC Z1.4-1993
What happens when AQL = . 1% isnt good
enough
AQL = 0.1% => 1000 ppm

Is Z1.4 Adequate?
How would you decide?
If not, what would you do?
Construct O-C curve for n=1000, c=0 (Poisson). Use
100ppm < p < 5000 ppm (see slides 38 & 39)
Acceptance Sampling Webinar
20101129

51

ANSI/ASQC Z1.9-1993
Mil-Std 414

Sampling for Variables; 110 page Document

Four Sections in the document
Section A: General description of Plans
Section B: Plans used when variability is unknown
(Std. deviation method is used)
Section C: Plans used when variability is unknown
(range method is used)
Section D: Plans used when the variability is known.
Acceptance Sampling Webinar
20101129

52

ANSI/ASQC Z1.9-1993
Mil-Std 414

Information necessary: N, AQL, Inspection Level

How to Use
Code Letters
Single or Double Limit, Std. Dev or Range Method Plans
Switching Rules
Obtain: Code Letter, n, Accept/Reject criteria,

critical statistic (k)

O-C Curves

Acceptance Sampling Webinar

20101129

53

ANSI/ASQC Z1.9-1993
Exercise (From QCI, CQE Primer, pVI-37)

The specified max. temp for operation of a device is

209F. A lot of 40 is submitted for inspection. Use
Normal (Level II) with AQL = 0.75%. The Std. Dev.
is unknown.
Use Std. Dev. Method, variation unknown
Find Code Letter, Sample Size, k
Should lot be accepted or rejected

Acceptance Sampling Webinar

20101129

54

Z1.9 Code Letters

For N=40, AQL=0.75 |||||| Use AQL=1.0 & Code Letter D

Acceptance Sampling Webinar
20101129

55

Z1.9 Finding Decision Criteria

Std. Dev method Table B-1

For Code Letter D, n=5 & AQL=1, k=1.52

Acceptance Sampling Webinar
20101129

56

ANSI/ASQC Z1.9-1993
What is k
k is a critical statistic (term used in hypothesis testing).
It defines the maximum area of the distribution which can be
above the USL.
When Qcalc > k, there is less of distribution above Qcalc than above
k and lot is accepted. (Compare to Z table)
Increasing (USL - X-bar) increases Pa

Acceptance Sampling Webinar

20101129

57

ANSI/ASQC Z1.9-1993
Exercise Solution

The five reading are 197F, 188F, 184F, 205F, 201F.

X-bar (mean) = 195F
S (Std. Dev) = 8.8F
Qcalc = (USL X-bar)/s = 1.59
Because Qcalc = 1.59 is greater than k=1.52, lot is

accepted

Acceptance Sampling Webinar

20101129

58

Z1.9 OC Curve for D

Table A-3 (p9)

Acceptance Sampling Webinar

20101129

59

ANSI/ASQC Z1.9-1993
Another Exercise
Same information as before
AQL = 0.1
Find Code Letter, n, k
Accept or Reject Lot?

Acceptance Sampling Webinar

20101129

60

Solution 2nd Exercise

New code letter is E, n=7, & k=2.22
The seven reading are 197F, 188F, 184F, 205F, 201F,
193F & 197F.
X-bar (mean) = 195F
S (std. Dev) = 7.3F
Qcalc = (USL X-bar)/s = 1.91
Because Qcalc = 1.91 is less than k=2.22, lot is
rejected

Acceptance Sampling Webinar

20101129

61

Inspection Economics
Average Total Inspection: The average number
of devices inspected per lot by the defined sampling
plan
ATI = n Pa + N(1- Pa)
which assumes each rejected lot is 100% inspected.
Average Fraction Inspected:
AFI = ATI/N
Average Outgoing Quality:
AOQ = AQL (1 AFI)
Acceptance Sampling Webinar
20101129

62

Inspection Economics
Exercise (from Grant & Leavenworth, p395)

AQL = 0.5%, N=1000

Which sampling plan would have least ATI.
n = 100, c = 0
n = 170, c = 1
n = 240, c = 2

Acceptance Sampling Webinar

20101129

63

Inspection Economics
Exercise Solution
N

1000

1000

1000

100

170

240

Pa

0.59

0.8

0.92

n Pa

59

136

220.8

N(1- Pa)

410

200

80

ATI

460

336

300.8

AFI

0.460

0.336

0.301

AOQ

0.0027 0.00332

Acceptance Sampling Webinar

20101129

.00349
64

Inspection Economics
Comparison of Cost Alternatives
No Inspection
NpD

100% Inspection
NC
Sampling
nC + (N-n)pDPa + (N-n)(1-Pa)C
D = Cost if defective passes; C = Inspection cost/item
Acceptance Sampling Webinar
20101129

65

Inspection Economics

Sample Size Break-Even Point

nBE = D/C
D = Cost if defective passes; C = Inspection cost/item

Acceptance Sampling Webinar

20101129

66

Resources
American Society for Quality
Quality Press
www.asq.org
ASQ/NC A&T partnership quality courses
CQIA, CMI, CQT, CQA, CQMgr, CQE, CSSBB
Quality Progress Magazine
And others
Web-Sites
www.stattrek.com excellent basic stat site
http://mathworld.wolfram.com/ - greaqt math and stat site
Acceptance Sampling Webinar
20101129

67