MDP 308 Quality Management 2020-02-09 · Select p values from 0 to an upper limit suitable to the...
Transcript of MDP 308 Quality Management 2020-02-09 · Select p values from 0 to an upper limit suitable to the...
MDP 308
Quality Management
Lecture #10
Acceptance Sampling
Today’s lecture
What is acceptance sampling?
Selecting items randomly for inspection
Problems with sampling
Single sampling plan and OC curves
Double sampling plan
Rectified sampling plan and AOQL
Using military standards for acceptance sampling
What is acceptance sampling?
Acceptance sampling is a method used to accept or reject a lot (batch) of products based on a random sample of the lot.
The purpose of acceptance sampling is to sentence lots (accept or reject) rather than to estimate the quality of a lot.
Acceptance sampling plans do not improve quality. The nature of sampling is such that acceptance sampling will accept some lots and reject others even though they are of the same quality.
The most effective use of acceptance sampling is as an auditing tool to help ensure that the output of a process meets requirements.
Problems with sampling
The sampling rules in any formal sample plans are based on probability. But application of probability predicts the acceptance of lots with substandard quality.
Example: A lot containing 100 items with a probability of an item to be defective of 5% (this is called the quality level of the lot). A sampling plan is decided such that 20 items will be drawn randomly, if the number of defective items is found to be zero or 1, the lot will be accepted; otherwise, the lot will be rejected. In this example, number of defective items = 0.05 x 100 = 5
The probability of accepting the lot =
Prob.{zero defective items are drawn from the 20-item sample}
+ Prob.{only one defective item is drawn out of the 20-item sample}
Problems with sampling
Using Hypergeometric distribution (sampling without
replacement), we have N = 100, n = 20, r = 100x5% = 5
The probability of accepting the lot =
Prob. {zero defective items are drawn from the 20-item
sample} + Prob.{only one defective item is drawn out of
the 20-item sample}
= +
= 0.42 + 0.319 = 0.739
5 100 5
0 20 0
100
20
−
−
5 100 5
1 20 1
100
20
−
−
Problems with sampling
Using Binomial distribution as an approximation (sampling
with replacement), we have N = 100, n = 20, p = 0.05
The probability of accepting the lot =
Prob. {zero defective items are drawn from the 20-item
sample} + Prob.{only one defective item is drawn out of
the 20-item sample}
= +
= 0.36 + 0.377 = 0.736
0200 )05.01(05.00
20−−
1201 )05.01(05.01
20−−
Acceptance sampling risks As mentioned, acceptance sampling can reject “good” lots and
accept “bad” lots. More formally:
Producers risk refers to the probability of rejecting a good lot. In order to calculate this probability there must be a numerical definition as to what constitutes “good” AQL (Acceptable Quality Level) - the numerical definition of a good lot.
The ANSI/ASQC standard describes AQL as “the maximum percentage or proportion of nonconforming items or number of nonconformities in a batch that can be considered satisfactory as a process average”
Consumers Risk refers to the probability of accepting a bad lot where: LTPD (Lot Tolerance Percent Defective) - the numerical definition of a
bad lot described by the ANSI/ASQC standard as “the percentage or proportion of nonconforming items or non-comformities in a batch for which the customer wishes the probability of acceptance to be a specified low value.
Selecting items randomly for inspection
It is extremely important to have random samples in any
acceptance procedure to help in avoiding systematic and
random errors.
Sampling plans must be devised to fit the packaging from the
incoming pieces.
Example: Consider an item delivered in 100 boxes, each box
contains 4 rows of 9 pieces each. The sampling plan needs to
decide randomly which box to draw from, then which row and
then finally which box in the row.
…………
100boxes
1 2 3
4 5 67 8 9
123
4
Selecting items randomly for inspection
Using the random digits generation function in the calculator, we can get three random digits from which we need only two to decide which box to draw from. The digits that represent the boxes are from 00 to 99.
Then another random digits are generated and one digit is used to decide the row number in the selected box. Here we need one digit from 0 to 3.
Finally, another random digit from 0 to 8 is generated to decide which piece will be drawn from the selected row.
It is important to show that if a random digit is found to be outside the defined range, it must be neglected and another random digits are generated until the selected digit lies inside the defined range.
Single (one-stage) sampling plan
From a lot of N items, select n items randomly
Let X be the number of defective items found in the
sample
If X > c, the lot is rejected
If X c, the lot is accepted
The probability of accepting the lot is calculated using
binomial distribution as follows:
𝑃𝑎 = 𝑃𝑟𝑜𝑏. 𝑋 ≤ 𝑐 =
𝑘=0
𝑐𝑛𝑘
𝑝𝑘(1 − 𝑝)𝑛−𝑘
Where p is the probability that an item is defective.
The Poisson distribution chart
The operating characteristic (OC) curve
Probability of acceptance
(Pa)
Percent defective (100p)
1.0
Drawing an OC curve To draw an OC curve, we need first to identify two parameters:
n: sample size
c: limit on the number of allowable defective items in a sample, beyond which the sample will not be accepted.
The steps of constructing OC curves are:
1. Select p values from 0 to an upper limit suitable to the selected n and c values (maximum is 0.99)
2. multiply each value by n and enter it in the table
3. make a percent column for p and mark it 100p
4. Start at the pn value on the Poisson distribution graph and go to the curve with the corresponding c value to get Pa. Or if the sample size is less than 20, use the cumulative binomial distribution formula for calculating Pa as stated earlier.
5. Record the Pa value on the table.
Note: if the sample size (n) is less than 20 units the binomial distribution is used to build the OC Curve otherwise the Poisson distribution is used as sufficient approximation.
Example of drawing an OC curve
Example: Draw an OC curve with (n = 200 and c = 6)
We first construct the following table using the procedure
described in the previous slide:
p np 100p Pa
0 0 0 1.00
0.01 2 1 1.00
0.015 3 1.5 0.97
0.02 4 2 0.89
0.025 5 2.5 0.76
0.03 6 3 0.61
0.04 8 4 0.31
0.05 10 5 0.13
0.06 12 6 0.05
0.07 14 7 0.01
0.08 16 8 0.00
0.09 18 9 0.00
Example of drawing an OC curve (cont.)
On the OC curve we the following terms are defined:
AQL: Acceptable Quality Level
IQL: Indifference Quality Level
LTPD: Lot Tolerance Percent Defective
Indecisive zone = LTPD - AQL
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
1.10
0 1 2 3 4 5 6 7 8 9 10
100p
Pa
AQL IQL RQL
Indecisive zone
Producer and consumer risks
0
0.2
0.4
0.6
0.8
1
1.2
Pro
ba
bil
ity o
f A
cce
pta
nce
100P
OC Curve
LTPDAQL
Producers Risk
Consumers Risk
Hypothesis testing in acceptance sampling
The null and alternative hypotheses of the single-sampling plan
can be stated as follows:
H1: p > p0
H0: p = p0
Where p is the true percentage defective in the population (lot)
p0 is the rejection limit = c / N
The probability of type I error (-error) is the probability that
the lot will be rejected while its true percent defective is
actually less than or equal p0. This is what is called producer
risk.
The probability of type II error (-error) is the probability that
the lot will be accepted while its true percent defective is
actually greater than p0. This is what is called consumer risk.
Producer and consumer risks
0
0.2
0.4
0.6
0.8
1
1.2P
rob
ab
ilit
y o
f A
cce
pta
nce
100P
OC Curve
LTPDAQL
Producers Risk
Consumers Risk
Producer risk = 1 – Probability of acceptance at the AQL
Consumer risk = Probability of acceptance at the LTPD
The effect of n and c on the OC curve
Pa
100p
1.0
n increases
Pa
100p
1.0
c increases
The effect of n and c on the OC curve
0
0.2
0.4
0.6
0.8
1
1.2
0 2 4 6 8 10
100p
Pa
Designing sampling plans
Operationally, three values need to be determined
before a sampling plan can be implemented (for single
sampling plans):
N = the number of units in the lot
n = the number of units in the sample
c = the maximum number of nonconforming units in the
sample for which the lot will be accepted.
Designing sampling plans
While there is not a straightforward way of determining
these values directly given desired values of the
parameters, tables have been developed. Below is an
excerpt of one of these tables.
Excerpt From a Sampling Plan Table with Producers Risk = 0.05 and Consumers Risk = 0.10
c LTPD/AQL n(AQL) n(LTPD)
0 44.89 0.052 2.334
1 10.946 .355 3.886
2 6.509 .818 5.324
3 4.89 1.366 6.68
4 4.057 1.97 7.992
5 3.549 2.613 9.274
6 3.206 3.286 10.535
7 2.957 3.981 11.772
8 2.768 4.695 12.996
9 2.618 5.426 14.205
Designing sampling plans
Pa
100p
1.0
The ideal OC curve
Given the choice from a set of OC curves, select the one that has the lowest c/n
ratio and smallest LTPD-AQL difference
Three alternatives for specifying sampling
plans
1. Producers Risk and AQL specified
2. Consumers Risk and LTPD specified
3. All four parameters specified
First two alternatives of sampling plan
preparation
For the first two cases we simply choose the acceptance
number and divide the appropriate column by the
associated parameter to get the sample size.
Example 1: Given a producers risk of .05 and an AQL of .015
determine a sampling plan
c = 1: n = .355/.015 ~ 24
c = 4: n = 1.97/.015 ~ 131
Example 2: Given a consumers risk of .1 and a LTPD of .08
determine a sampling plan
c = 0: n = 2.334/.08 ~ 29
c = 5: n = 9.274/.08 ~ 116
Sampling plans when All four parameters
specified
When all four parameters are specified we must first find
a value close to the ratio LTPD/AQL in the table. Then
values of n and c are found.
Example: Given producers risk of .05, consumers risk of
.1, LTPD 4.5%, and AQL of 1% find a sampling plan.
Since 4.5/1 = 4.5 is between c= 3 and c = 4. Using the n(AQL)
column the sample sizes suggested are 137 and 197
respectively. Note using this column will ensure a producers
risk of .05. Using the n(LTPD) column will ensure a consumers
risk of .1
Double sampling plans
In an effort to reduce the amount of inspection, double
(or multiple) sampling is used. Whether or not the
sampling effort will be reduced depends on the defective
proportions of incoming lots. Typically, four parameters
are specified:
n1 = number of units in the first sample
c1 = acceptance number for the first sample
n2 = number of units in the second sample
c2 = acceptance number for both samples
Procedure of double sampling plans
A double sampling plan proceeds as follows:
A random sample of size n1 is drawn from the lot.
If the number of defective units (say d1 ) c1 the lot is
accepted.
If d1 c2 the lot is rejected.
If neither of these conditions are satisfied a second
sample of size n2 is drawn from the lot.
If the number of defectives in the combined samples (d1 +
d2) > c2 the lot is rejected. If not the lot is accepted.
Example
Consider a double sampling plan with n1 = 50; c1 = 1; n2 =
100; c2 = 3. What is the probability of accepting a lot with
5% defective?
The probability of accepting the lot, say Pa, is the sum of
the probabilities of accepting the lot after the first and
second samples (say P1 and P2)
Example (cont.)
To calculate P1 we first must determine the expected
number defective in the sample (50*.05)
we can then use the Poisson function to determine the
probability of finding 1 or fewer defects in the sample,
e.g., POISSON.DIST(1,2.5,TRUE) = .287
To calculate P2 we need to determine first the scenarios
that would lead to requiring a second sample. Then for
each scenario we must determine the probability of
acceptance. P2 will then be the sum of the probability of
acceptance for each scenario.
Example (cont.)
For this example, there are only two possible
scenarios that would lead to requiring a second
sample:
Scenario 1: 2 nonconforming items in the first sample
Scenario2: 3 nonconforming items in the first sample.
The probability associated with these two scenarios
is found as follows:
P(scenario 1) = POISSON.DIST(2,2.5,FALSE) = .257
P(scenario 2) = POISSON.DIST(3,2.5,FALSE) = .214
Example (cont.) Under scenario 1 the lot will be accepted if there are either 1 or 0
nonconforming items in the second sample. The expected number of nonconforming items in the second sample .05(100) = 5
the probability is then: POISSON.DIST(1,5,TRUE) = .04
therefore the probability of acceptance under scenario 1 is (.257)*(.04) = .01
Under scenario 2 we will accept the lot only if there are 0 nonconforming items in the second sample, i.e., POISSON.DIST(0,5,FALSE) = .007
therefore the probability of acceptance under scenario 2 is (.214)*(.007) = .0015
so the probability of acceptance on the second sample is .01 + .0015 = .0115
So the overall probability of acceptance is (probability of acceptance on first sample + probability of accepting after 2 samples) = .287 + .0115 = .299
OC curve for double sampling
OC Chart - Double Sampling Plan
0
0.10.2
0.3
0.40.5
0.6
0.7
0.80.9
1
0
0.0
1
0.0
2
0.0
3
0.0
3
0.0
4
0.0
5
0.0
6
0.0
7
0.0
7
0.0
8
0.0
9
0.1
0.1
1
0.1
1p
Pa
0
0.10.2
0.3
0.4
0.50.6
0.7
0.80.9
1
P1
Pa = P1 + P2
P(rejection 1st)
Average Sample Number
Since the goal of double sampling is to reduce the inspection effort, the average number of units inspected is of interest. This is easily calculated as follows
Let PI = the probability that a lot disposition decision is made on the first sample, i.e., (probability that the lot is rejected on the first sample + probability that the lot is accepted on the first sample)
Then the average number of items is:
n1(PI) +(n1 + n2)(1 - PI)
= n1 + n2(1 - PI)
The average sample number curve can be
derived from the data on the OC chart
ASN Curve
0
20
40
60
80
100
120
0
0.0
1
0.0
2
0.0
2
0.0
3
0.0
4
0.0
4
0.0
5
0.0
6
0.0
6
0.0
7
0.0
8
0.0
9
0.0
9
0.1
0.1
1
0.1
1
0.1
2
p
Av
era
ge
sa
mp
le n
um
be
r
Rectified sampling plan
Population
p, N
Sample
n
Reject
Prob. = 1-Pa Accept
Prob. = Pa
Average number of defectives = np < c
replace np defectives
N
n zero defectives p(N-n) defectives
= p(N-n) defectives
Make 100% inspection for N
replace for Np defectives
N Zero defectives
Rectified sampling plan
Expected number of defectives in the lot = (1-Pa)zero + Pa[zero+p(N-
n)] = pPa(N-n)
→ expected percentage defective in the lot = 100 pPa(N-n)/N
The average outgoing quality (AOQ) = pPa(1-n/N)
0
0.002
0.004
0.006
0.008
0.01
0.012
0.014
0.016
0.018
0.02
0 2 4 6 8 10
p %
AO
Q
Average outgoing quality limit (AOQL)
Military Standard 105E
(MIL STD 105E)
(ANSI/ASQC Z1.4, ISO 2859)
Most widely used acceptance sampling system for
attributes
MIL STD 105E is Acceptance Sampling System
collection of sampling schemes
Can be used with single, double or multiple
sampling plans
Inspection Types
Normal Inspection
Used at start of inspection activity
Tightened Inspection Instituted when vendor’s recent quality history has deteriorated
Acceptance requirements for lots are more stringent
Reduced Inspection Instituted when vendor’s recent quality history has been exceptionally
good
Sample size is usually smaller than under normal inspection
Switching Rules
- Production Steady
- 10 consecutive lots accepted
- Approved by responsible
authority
NormalReduced Tightened
- Lot rejected
- Irregular production
- Lot meets neither accept
nor reject criteria
- Other conditions warrant
return to normal inspection
2 out of 5 consecutive lots
rejected
5 consecutive
lots accepted
10 consecutive lots remain
on tightened inspection
Start
Discontinue
Inspection
AND conditions
OR conditions
Procedure for
MIL STD 105E
STEP 1: Choose AQL MIL STD 105E designed around Acceptable Quality Level,
AQL Recall that the Acceptable Quality Level, AQL, is producer's largest
acceptable fraction defective in process
Typical AQL range: 0.01% AQL 10%
Specified by contract or authority responsible for sampling
STEP 2: Choose inspection level
Level II
Designated as normal
Level I
Requires about one-half the amount of inspection as Level II
Use when less discrimination needed
Level III
Requires about twice as much inspection as Level II
Use when more discrimination needed
S-1, S-2, S-3, S-4
Four special inspection levels used if very small samples necessary
Procedure for
MIL STD 105E
STEP 3–Determine lot size, N
Lot size most likely dictated by vendor
STEP 4: Find sample size code letter
From Table 12-10
Given lot size, N, and Inspection Level, use table to
determine sample size code letters
STEP 5: Determine appropriate type sampling plan
Decide if Single, Double or Multiple sampling plan is to be
used
Procedure for
MIL STD 105E
STEP 6: Find Sample Size, n, and Acceptance Level, c
Given sample size letter code, use Master Tables: 12-11, 12-12,
and 12-13
Find n and c for all three inspection types:
Normal Inspection
Tightened Inspection
Reduced Inspection
Procedure for
MIL STD 105E
Example
Suppose product comes from vendor in lots of size 2000 units. The acceptable quality level is 0.65%. Determine the MIL STD 105E acceptance-sampling system.
Normal Insp. Level
Lot Size
= 2000
Table 12-10
AQL
Plan K
Sample 125 units.
Ac = 2, accept if defects ≤ 2.
Re = 3, reject entire lot if defects ≥ 3.
Table 12-11
AQL
Plan K
Sample 125 units
Ac = 1, accept if defects ≤ 1.
Re = 2, reject entire lot if defects ≥ 2.
Table 12-12
AQL
Plan K
Sample 50 units
Ac = 1, accept if defects ≤ 1.
Re = 3, reject entire lot if defects ≥ 3.
Table 12-13