Management of Quality

Section C - Acceptance Sampling

1. Acceptance Sampling Inspection is often used to accept or reject incoming batches of raw materials or products, to stop faulty material causing problems and non-conformance. Frequently the number of items requiring inspection is so great that time is not available, or the cost of 100% inspection would be prohibitive. The potential cost of no inspection could also be very great. Acceptance Sampling techniques use statistical and probability theory to allow a decision to be made on the basis of a random sample taken from the batch. Advantages Saves time and money Ideal for destructive testing (e.g. bottled wine) Less handling to cause damage (e.g. fruit) Rejecting entire batches motivates supplier! Disadvantages Uncertainty always exists Complex - requires planning and documentation Concepts contradict modern TQM approach 2. Single Sampling Plan for Attributes

Take a random sample of size n Inspect each item independently If more defectives are found than c, the Acceptance Number, reject the whole batch, otherwise accept.

A single sampling plan is defined by n and c. The Binomial Distribution describes the probability of obtaining r defectives from a sample of n.

c n! P(Acceptance) = pr q n r r !( n r )! r =0

The probability of acceptance for given c can be plotted against p. This is called the Operating Characteristic (OC) curve of the sampling scheme.

Acceptance Sampling

2. Operating Characteristic Curves

OC Curve s for Single Sa m pling Schem e s

1.00000 0.90000 0.80000 0.70000 Probability of Acceptance 0.60000 0.50000 0.40000 0.30000 0.20000 0.10000 0.00000 0.03 0.06 0.09 0.00 0.01 0.02 0.04 0.05 0.07 0.08 0.1 0.11 0.12 0.13 0.14 0.15

Batch Pr opo rtio n De fe ctive P(a)50,3 P(a) 100,3 P(a) 200,3 P(a) 300,3

O C courve s for single sa m pling sche m e s

1.2000

AQL = 0.06
1.0000 Probability of Acceptance p(a) 0.8000 0.6000 0.4000 0.2000 0.0000 0.01 0.02 0.08 0.1 0.11 0.12 0.13 0.00 0.03 0.04 0.05 0.06 0.07 0.09 0.14 0.15

Ideal OC curve always accepts or rejects at desired batch P

Batch p r opo r tion De fe ctive (P)

p(a) 100,2

p(a) 100,3

p(a) 100,4

p(a) 100,5

Acceptance Sampling

3. Definitions Acceptance Number (c)- the maximum number of defective items which is acceptable in a defined sample size. Acceptable Quality Level (AQL)- The proportion defective in the batch which is the worst level the consumer is prepared to accept as a process average. or the proportion defective for which it is desired to accept all batches. Lot Tolerance Proportion (Percent) Defective (LTPD) - the poorest proportion defective the consumer will accept for any individual lot (batch). or the proportion defective for which it is desired to reject all batches. Average Outgoing Quality (AOQ) - The average outgoing proportion defective for batches subject to a particular sampling plan. Average Outgoing Quality Level (AOQL) The worst level of AOQ for a particular sampling scheme, as incoming proportion defective varies. 4. Acceptable Quality Level (AQL) and Lot Tolerance Proportion Defective (LTPD)
OC Curve of Single Sampling Plan, n=100, c = 2
1.00000 0.90000 Probability of Acceptance 0.80000 0.70000 0.60000 0.50000 0.40000 0.30000 0.20000 0.10000 0.00000 0.00 0.06 0.07 0.01 0.02 0.03 0.04 0.05 0.08 0.09 0.1

Producers Risk

AQL

LTPD

Consumers Risk

Batch Proportion Defective

Producers Risk is a Type I Error - the probability that a good batch (better than the AQL) will be rejected. Consumers Risk is a Type II Error - the probability that a bad batch (worse than the LTPD) will be accepted. 5. Designing a Single-sample Sampling Scheme
Acceptance Sampling 3

Supplier and customer agree on batch size N, required AQL and LTPD, and . E.G. N = 1000 AQL = 0.01 = 0.05 LTPD = 0.045 = 0.10 OC should pass through (AQL,) and (LTPD,) A single-sampling scheme is completely specified given values for N, n and c. Since these are integers, the chance of passing exactly through the 2 points is small, and the simultaneous equations involved are non-linear. Tables or nomograms based on the cumulative Binomial or Poisson distributions are used to obtain alternative schemes. These are found in a variety of Sampling Standards and textbooks. Answers n = 137, c = 3 n = 149, c = 3 n = 197, c = 4 n = 178, c = 4 > = 0.050, = 0.134 > = 0.065, = 0.100 > = 0.050, = 0.058 > = 0.043, = 0.100

Acceptance Sampling

Average Outgoing Quality

0.020 0.019 0.018 0.017 0.016 Average Proportion Defective of Outgoing Lots 0.015 0.014 0.013 0.012 0.011 0.010 0.009 0.008 0.007 0.006 0.005 0.004 0.003 0.002 0.001 0.000 0.00 0.03 0.04 0.06 0.07 0.08 0.1 0.11 0.14 0.15 0.01 0.02 0.05 0.09 0.12 0.13

AOQL

Incoming Lot Quality

6. Types of Sampling Scheme a) Single Sampling. A single sample is taken from the batch, and the disposition of the batch depends only on the information contained in that sample. b) Double Sampling. A sample is inspected from the batch, and either: The batch is passed Another sample is taken The batch is rejected c) Multiple Sampling. Similar to double sampling, but more than two samples may be required to reach a decision on batch disposition. d) Sequential Sampling. Small samples or individual items are inspected until a result is obtained. Sampling schemes of all these types can be designed to give equivalent OC curves. Single sampling schemes are simplest, but potentially involve more inspection for clear cut batch decisions. 7. Double-Sampling Schemes Double-Sampling schemes are defined by the following parameters both first and second samples. n1 and n2 are sample sizes c1 and c2 are acceptance numbers Re1 and Re2 are rejection numbers

Acceptance Sampling

The first random sample is usually small, to filter out the obviously poor batches. The batch is accepted immediately if r c1, and rejected if r > Re1. If r is between c1 and Re1, a second sample is taken. n2 is usually larger. Now the combined number of defectives obtained in the two samples is used to determine the batch disposition, using Re2 and c2, where Re2 = c2 + 1. The average amount of inspection is smaller than with a single sampling scheme. NB. sample inspection may be curtailed if c is exceeded for any scheme. 8. Multiple Sampling Plans Multiple sampling plans have a number of cumulative stages, in which small samples are taken. Scheme parameters are n1...ni, c1....ci, Re1....Rei Example A simple seven stage scheme
Sample 1 2 3 4 5 6 7 n 10 10 10 10 10 10 10 n 10 20 30 40 50 60 70 c 0 1 3 4 5 7 10 Re 4 5 7 8 9 10 11

c and Re converge on the AQL (p 0.15 or 15%)

12 10 8 c & Re 6 4 2 0 1 2 3 4 S a m p le N u m b e r

Reject

Accept

9. Acceptance Sampling - Advantages of Different Schemes Advantage Psychological Acceptability Amount of Inspection Admin/Training Single Only one chance Greatest Least Double Adequate Often 10 15% less More Multiple Indecisive & complex Often even less Greatest

10. Switching to Tightened and Reduced Inspection

Acceptance Sampling

The effectiveness of ongoing sampling inspection is increased by switching between different schemes, depending on the current incoming quality level. When incoming quality deteriorates, inspection switches to tightened scheme according to firm rules. Some standards also make provision for reducing the level of inspection. Example (From MIL-STD 105D/ BS 6001 see Kehoe p123 et seq.) Batch Size 35001 to 150000 AQL 0.25% (p = 0.0025) Single sampling Scheme n c Reduced 200 1 Normal 500 3 Tightened 500 2

Re 4 4 3

N.B. Reduced inspection may lead to ambiguity. Batch is accepted, but normal inspection reinstated with next batch.
Tightened and Reduced Inspection 0.03
Reduce

0.02 AOQ
Norm al

0.01
Tightened

0.00 0.00 0.04 0.08 Incoming Quality Level 0.12

11. Average Sample Number (ASN) The sample number for a single-sampling scheme is fixed. ASN for a double or multiple scheme varies according to the probability of accepting and rejecting first and subsequent samples. Curtailment will also reduce the ASN.

Acceptance Sampling

Average Sample Number

140 120 100

ASN

80 60 40 20 0 0 0.04 0.08 0.12

Batch Proportion Defective

ASN curve for a double sampling scheme n1 = 60, n2 =120, c1 = 0, Re1 = 3, compared with single scheme n= 100. (NB. Double sampling is sometimes more expensive.) 13. Average Total Inspection (ATI) For a single-sampling plan with rectification:-

A T I= n + (1 P a)( N n)

Where Pa = probability of acceptance of a batch having a specified proportion defective.

Average Total Inspection 1200 1000 800 ATI 600 400 200 0 0.00 0.02 0.04 0.06 0.08 0.12 Batch Proportion Defective p 0.14 0.1

ATI curve for N = 1000, n = 100, c=2. 14. Zero Acceptance No Zero Defects Lot-by-lot attribute sampling plans: All product to one customer Supplied in lots

Acceptance Sampling

Sample inspected Accept if c or fewer non-conformances found

Assumptions

process usually stable an AQL agreed some non-conforming items expected

Zero Defects for process or for individual batch is not ensured by making c = 0. Some probability of accepting batches with NC product always

1.0 0.9 0.8

Probability of Acceptance

Single Sampling - Acceptance No (c) = 0

0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 .00 .04 .10 .13 .01 .02 .03 .05 .06 .07 .08 .09 .11 .12 .14
Batch Proportion D efective

P (a)100,0

P (a)100,2

P (a) 100,3

P (a) 100,4

P (a) 100,5

OC curves for c= 0 are always convex Confidence Level Approach. If Acceptance Number is to be zero, a simple method for calculating the required sample size for a given confidence level uses the following formula:-

n>

Log 1 C Log 1 p

Where 100C% is confidence level and p is LTPD E.g. C = 0.95 and p = 0.01

n>

Log 1 0. 95 Log 1 0. 01

n > 298.1 (say 300)

Acceptance Sampling

.15

By inspection of 300 items we can be 95% confident that not more than 1% of items are NC if no defectives have been found. 15. Acceptance Sampling using Variables Using measurement, more information can be obtained, and hence the same protection (i.e. equivalent OC curve) can be achieved with a smaller sample, which is useful for very low AQLs. More information can also help with process improvement and control. Measurement however may cost more per item than the collection of attribute data, and attribute plans can cover several different quality characteristics. Distribution of the variable quality characteristic must be normal. Note: It is sometimes possible to reject a batch when no non-conforming items have been found! a) Main Types of Variable Sampling Scheme Proportion non-conforming Process Mean

b) Variables Sampling Plans Sampling plans are based on: - known or must be estimated from sample s. Single (i.e. upper or lower) or double spec limits. k value found from tables depending on AQL. For double spec limits the AQL may be set separately for each of the limits. Shipment size. Procedure: Calculate Z for each spec limit. i.e. for single USL

z = USLs x
If Z is too small, then it is likely that the lot x and/or s has changed. Reject the lot. To decide if Z is too small, compare it with a k value from tables chosen to match the agreed AQL. Example 1. Using MIL STD 414 ( unknown) Find code letter for the batch size from table (Montgomery Table 14.1 p 632). Using Table 14.2 read sample size from the first column, together with a value of k for the appropriate AQL. E.g. A shipment size of 2000 and an inspection level of IV gives code L from Table 14.1 . Table 14.2 for normal inspection with AQL of 2.5% gives sample size of 40 and k = 1.58. We have a single lower specification limit - i.e. not more than 2.5% below 263.4 units 1. 2. Take a sample of 40 and find x and s Calculate Z using:

ZL=

x LSL s

where LSL = 263.4 the lower specification limit 3. Accept the shipment if Z > 1.58, reject if Z< k
Acceptance Sampling 10

16. Variables Plan for Sample Means ( known) Design Procedure - Single Spec Limit 1. Determine

AQL

and
m

LTPD

for sample mean, plus and the Type I and Type II errors.

2. Calculate limit

for sample means, as follows:-

X
and

=
=

X
X

AQL

+ Z
+ Z

for AQL batches n

n
for LTPD batches

X
Note:-

LTPD

is negative.

3. Solving these equations simultaneously we can obtain n, the sample size, and limiting value of the sample mean. Example 2

, the

A towing rope must have a breaking load of at least 4000kg. It is known that the distribution of breaking loads follows a normal distribution with = 250 kg. Design a single sampling scheme to control the sample mean of incoming batches, given that AQL is 4500kg ( = 0.05) and the LTPD is 4250kg ( = 0.10).

X
and

AQL

+ Z

250 = 4500 + 1. 645 n n

LTPD

+ Z

250 = 4250 + 1. 28 n n

Equating and solving gives:Sample size n = 8.555 and

X = 4359.4kg
m

Answer:- Draw a sample of 9 test ropes, and compare mean breaking load with the batch if sample mean is less than X m . Relationship of X m to Sample mean for Batches of:-

X . Reject
m

Acceptance Sampling

11

= 0.10

= 0.05
xLTPD = 4250
LTPD Quality 16. Acceptance Sampling Plans - Summary a) Acceptable Quality Level (AQL) Methods Application - when an AQL of defective product has been agreed between supplier and customer, and product is supplied in a stream of lots or shipments. Provides confidence that the process is running at or better than the agreed AQL. e.g. BS 6001 (1991), MIL STD 105D (Attributes), MIL STD 404 & ISO3951 (Variables). b) Average Outgoing Quality Level.(AOQL) Methods Application - as for AQL methods, to provide assurance that the overall level of defective product produced is less than the established AOQL. e.g. Dodge-Romig plans and other attribute sampling schemes c) Lot Tolerance Percent Defective (LTPD) Methods Application - For use when protection against receipt of particularly bad lots (worse than LTPD) is required, either in a stream of lots or for isolated shipments. e.g. Dodge-Romig plans and other attribute sampling schemes

xAQL = 4500

AQL Quality

Acceptance Sampling

12

Example Sheet 3 - Acceptance Sampling

1. a) You have negotiated with a supplier for regular shipments of 8000 capacitors to be delivered each month. The agreed AQL is 2.5% at Inspection Level II, using MIL STD 105D (relevant tables attached). What sample size(s) and acceptance number(s) would you use in (normal inspection level) attribute sampling schemes for:i) Single inspection ii) Double inspection b) Devise an alternative scheme using a zero acceptance number, if the LTPD is 5% defective, and the required confidence level is 95%. 2. a) At a canning plant, an acceptance sampling scheme is in use for incoming batches of soft fruit. The scheme is a single-sampling attribute plan, in which random samples of 80 are taken and inspected for defects. If more than 4 items of fruit are found non-conforming to the specification standards, the entire batch is rejected (N.B. a binomial situation). Using a statistical package, spreadsheet, or manually, graph the OC curve for this sampling scheme. b) As the fruit deteriorates rapidly, the normal practice is to immediately 100% inspect and sort any rejected batches, using only the good items for canning. Estimate the average number of staff needed to undertake this task if 5 batches of 5000 items each with a proportion defective expected to be 5%, arrive each day in the picking season. Assume a person can examine 15 items/minute, and works 7 hours each day. 3. The diameter of a car wheel spindle is specified as 1.500 in 0.0025 in. From process capability studies, the spindle manufacturers have determined that the process standard deviation is 0.0005 in, and that the distribution of diameters is normal. (You note that this gives a Cp of 1.667, which implies that the distribution cannot significantly overlap both tolerance limits simultaneously, although if the process mean is not centered it might overlap either one, causing non-conformance.) You wish to source large batches of spindles from this manufacturer, and have agreed on a variables sampling scheme with AQL of 2% defective (0.020), with a probability of acceptance of 0.95 ( i.e. = 0.05); and LTPD of 3.5% defective (0.035) with a probability of acceptance = 0.10. a) Using tables of the normal distribution in reverse, determine the ZAQL and ZLTPD values which would leave 0.020 and 0.035 respectively of the population of spindles outside of the lower specification limit. Hence obtain the expected sample means - X AQL for batches of AQL quality, and X LTPD for LTPD quality. b) As both upper and lower specification limits are involved, obtain Z/2 and Z/2 from tables in a similar way, and write down equations for X m using both values. Hence obtain the sample size required. c) Calculate the lower limiting value of sample mean for the shaft diameters, and by symmetry, the upper value.

Acceptance Sampling

13

Solution to Example Sheet 3 - Acceptance Sampling

Answers
1. a) Using BS6001/MIL STD 105D tables. For a batch size of 8000 and General Inspection Level II; code letter = L. i) From Table II-A, for AQL 2.50%, sample size is 200 and c = 10. ii) From Table III-A, first sample size = 125, with c = 5 and Re = 9, second sample size = 125, with c = 12 and Re = 13 c) n > n> Log(1-C) Log(1-p) Log(1- 0.95) = 58.4 Log(1 - 0.05)

For c= 0 a sample size of 59 will provide the necessary confidence. 2. a) Using the Excel spreadsheet, the OC curve is as follows.

P(a) 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 0 0.01

OC Curve for Fruit Sampling Scheme

0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0.11 0.12 0.13 0.14 0.15 Proportion Defective in Batch

b) Average Total Inspection (ATI) is given by n + (1 - Pa)(N-n). From OC curve, probability of acceptance Pa at 5% defective is 0.63. ATI = 80 + (1- 0.63)(5000 - 80) = 1900.4 for each batch. There are expected to be 5 similar batches each day, hence the total inspection is likely to be 5 1900.4 = 9502 items. One person can inspect/sort 15 60 7 = 6300 items of fruit per day. hence two people will be required to meet the expected workload. 3. Assuming upward shifts in process mean are as likely as downward shifts. a) Using Normal tables in reverse, ZAQL is found where 2% (0.02) of the distribution is in the tail, i.e.

enter table with (0.5 - 0.02) = 0.4800.

Acceptance Sampling

14

ZAQL = 2.05. Similarly ZLTPD = 1.81 Hence, as we know the population mean, For batches of AQL quality, X AQL = LSL + ZAQL = 1.4975 + 2.05(0.0005) = 1.498525 For batches of LTPD quality, b) = 0.05 and = 0.10. Z/2 and Z/2 are obtained from tables by entering with (0.5 - /2) = 0.4750 and (0.5 - /2) = 0.4500 respectively.
Z/2 = -1.96 and Z/2 = 1.645 For AQL batches and for LTPD batches

LTPD

= LSL + ZLTPD = 1.4975 + 1.81(0.0005) = 1.498405

=
=

AQL

+ Z /2

n
n
(1) AQL quality

LTPD

+ Z /2

Z/2 = -1.96 =

1. 498525

0. 0005 / n
m

Z/2 = 1.645 =

1. 498405

0. 0005 / n

(2)

LTPD quality

Subtracting (1) from (2) and expressing in terms of n

n=

3. 605 0. 0005 0. 00012

Sample size n = 15.02082 = 225.625 226. c) Substituting in either (1) or (2), shifts.

= 1.49846, which is the limiting value of sample mean for downward

Upward shift limit is given by 1.5000 + (1.5000 - 1.49846) = 1.50154 Hence the limiting values of shaft diameter sample means to accept the batch are: 1.49846 < sample mean < 1.50154.

Acceptance Sampling

15