33

STATISTICAL QUALITY CONTROL

33.1 INTRODUCTION
We have studied the definition and meaning of quality in the earlier chapter on TQM. Quality is mainly
what customer wants. By quality, an industrial engineer and an operations manager mean meeting
the customer requirements, both in product and services. The area of quality control is very old in our
literature. Statistical quality control methods were introduced by Walter Shewhart in a Bell Laboratory
memorandum in 1924. The enlarged and refined concepts of quality control and quality assurance got
the acceptance in industry very quickly. VlIe have also seen in the previous chapter that statistical-quality-
control charts are one of . the seven QC tools in the implementation of TQM. Let us discuss process
control charts (for quality) in more details. Then, we will• discuss the quality assurance and operating
characteristics curves.

33.2 PROCESS CONTROL

When we consider products or services, certain amount of variations are commonly present in the quality
characteristics which customer wants. These variations are due to two reasons:
(i) Chance variations.
(ii) Variations due to assignable causes.
Chance variations occur in random manner and there is very little that we can do about them to
control. These are natural to the process. It is due to aggregation of many relatively small factors present
in small degree. Few examples of chance variations are: quality of raw materials, performance of operator,
minor changes in weather conditions like temperature, etc. These variations are essentially uncontrollable,
inherent, random, chance-governed and commonly present. If the process shows only such variations,
it is said to be in-control (i.e., statistically controlled).
A different type of variations are due to some assignable cause. These variations are not random
and are relatively larger. It is generally caused by only one factor and occasionally by a few factors.
Some examples are: poor quality raw material from a new supplier, poor surface finish due to worn-
out tool, variation in the length of work-piece due to poor setting of machine, etc. These variations
are sporadic, uncontrollable, assignable and specific. Once, these variations are present, process is called
out-of-control.
512 INDUSTRIAL ENGINEERING AND MANAGEMENT

Chance variations cause about 85% of the problems in process, while assignable variations account
for only 15% of the problems. Chance variations may be reduced by changing the process and this can
be done through management efforts. Therefore, it is the management responsibility to control the chance
variations. To do so, there may be more efforts needed in better machine procurement, better training,
better environment and temperature control, etc. Assignable variations may be reduced by controlling
the process with proper identification and rectification of causes. Workers, who are closer to machines
(as compared to managers), can understand the causes and rectification of assignable variations. Hence,
it is workers' responsibility to control the assignable causes. This can be achieved by first identifying
the variations in the product/service parameters, relating them to the causes in the process variation or
process inputs and finally solving them by proper rectification measures. For the identification of assignable
variations, we use control charts.
33.3 CONTROL CHARTS
Control charts are the plots used to monitor outputs or inputs of a process, which produces a product
or service for meeting the customer's requirements. When we use control charts to monitor the processes,
it is called as statistical process control (SPC).
Two types of control charts are commonly used. These are shown in Figure 33.1.

[ Control Charts
...w.

ControlI
Charts Control Charts
for Variables for Attributes

R a or S
Chart Chart Chart

Figure 33.1 Classification of Control Charts

Variable data are the measured value of a product (or service) characteristics, such as length, weight,
diameter, temperature, hardness, etc. Attributes data are mainly of Go-No-Go type, proportion or fraction
of defective (or deviation) type. In this category, P-chart uses fraction defective of good/bad type,
C-chart uses number of defectives per unit (when a single defective may not be of greater significance
but a large number of defectives 'would add up to a defective item). A detail consideration for all the
charts are given in Table 33.1 and Table 33.4. In these tables, the formula for control limits of the
chart is also given. The values of constants in the formula (for example, A2 in X chart, D4 and D3 in
R chart, A3, B4 and B3 in X and S chart) are given in Table 33.3, for, different Values of sample size
(n).
STATISTICAL QUALITY CONTROL 513

Table 33.1 Comparison of Variable and Attribute Charts

Features Variable Charts Attribute Charts

Data required Variables such as measurable length, Attributes such as number of misprints
width, diameter, time, temperature, • per page, number of defects per sample
etc. etc.
Types of Charts R and S (or, a) charts P, NP, C and U charts
Advantages (i) Utilizes information from the (i) Utilizes information ftom inspection
inspection records. record.
(ii) Process average and variations (ii) Simpler than X and R Charts
are indicated.
(iii) Provides overall picture of quality.
Limitations (i) Proper training is needed to (i) Oetailed information for controlling
. follow the charts. a specific characteristic is lacking.
(ii) Control limits and specification (ii) Charts fail to recognize different types
limits are often mixed, which may of defects in one measurement or
cause confusion counting.
(iii) Inspection, using gauges of Go-
No-Go or counting of defects per
unit, does not fit in these charts.
Applications For the process control to check For the process contral to check
variations in the individual variations in proportion of defectiveg
characteristics of output or input or number of defects per unit
of the process

Table 33.2 Variable Control Chart

Features X and R Chart X and S Chart

Purpose X-Chart: To control central tendencies X-Chart: To control central tendencies
R-Chart: To control process variability S Chart: To control process variability
Control Limits X Chart X-Chart
UCL, =x + A2 R UCL,. = )7+4 g
LCL, = + A2 I? LCL, =x + A3 S
R-Chart S-Chart
UCLR = D4 k UCLS = B4 S
LCLE, = D3 R LCLS = B3 S
Sample Size R-Chart for
._
small ,sub-group. Sample (< 8) S-Chart for large sample size.
Central Line Mean of X and R Mean of X and. S

X= Viz: for n samples of mean X1 X = Ekiht: for II samples of mean X1

R= ER/. for e samples of range R. S = E S,/n: for e samples of standard

deviation S,
Use Measurement of quality characteristics Measurement of quality characteristics
which are variable like diameter, length, which are variable like diameter, length,
weight, etc. weight, etc.
514 INDUSTRIAL ENGINEERING AND MANAGEMENT

Table 33.3 Factors for Computing 3u Control Limits

Number Chart for Averages Chart for Standard Deviation Chart for Ranges
of obser- Factors for Control Limits
Factors for Factors Factors for Factors
vations Control Limits for Control Limits for
in Sample Central Central
Line Line

(n) A A, A2 c2 8, B2 83 84 d2 03 Di D2 D3 D4

2 2.121 3.760 1.880 0.5642 0 1.843 0 . 3.267 1.128 0.853 0 3.686 0 3.267
3 1.732 2.394 1.023 0.7230 0 1.858 0 2.256 1.693 0.888 0 4.358 0 2.575
4 1.500 1.880 0.729 0.7979 0 1.808 1) 2.266 2.059 0.880 0 4.698 0 2.282
5 1.342 1.596 0.577 0.8407 0 1.756 0 2.089 2.236 0.864 0 4.698 0 2.115
6 1.225 1.411) 0.483 0.8686 0.026 1.711 0.030 1.970 2.234 0.848 0 5.018 0 2.004
7 1.134 1.277 0.419 0.8882 0.105 1.672 0.118 1.882 2.704 0.833 0.205 5.203 0.076 1.924
8 1.061 1.175 0.337 0.9073 0.167 1.638 0.185 1.815 2.847 0.820 0.387 5.307 0.136 1.864
9 1.000 1.094 0.337 0.9139 0.219 1.609 0.239 1.761 2.970 0.808 0.546 5.394 0.184 1.816
10 0.949 1.028 0.308 0.9227 0.262 1.584 0.284 1.716 3.078 0.797 0.687 5.469 0.223 1.777
11 0.905 0.973 0.285 0.9300 0.299 1.561 0.321 1.679 3.173 0.787 0.812 5.534 0.256 1.744
12 0.866 0.925 0.266 0.9359 0.331 1.541 0.354 1.646 3.258 0.778 0.924 5.592 0.284 1.716
13 0.832 0.884 0.249 0.9410 0.359 1.523 0.382 1.618 3.336 0.770 1.026 5.646 0.308 1.692
14 8.802 0.848 0.235 0.9453 0.384 1.507 0.406 1.594 3.407 0.762 1.121 5.693 0.329 1.671
15 0.775 0.816 0.223 0.9490 0.406 1.492 0.428 1.572 3.472 0.755 1.207 5.737 0.348 1.652
16 0.750 0.788 0.212 0.9523 0.427 1.478 0.448 1.552 3.532 0.749 1.285 5.779 0.364 1.636
17 0.728 0.762 0.203 0.9551 0.445 1.465 0.466 1.534 3.588 0.743 1.359 5.817 0.379 1.621
18 0.707 0.738 0.194 0.9576 0.461 1.454 0.482 1.518 3.640 0.734 1.426 5.854 0.392 1.608
19 0.688 0.717 0.187 0.9599 0.477 1.443 0.497 1.503 3.689 0.733 1.490 5.888 0.404 1.596
20 0.671 0.697 0.180 0.9619 0.491 1.433 0.510 1.490 3.735 0.729 1.548 5.922 0.414 1.589
21 0.655 0.679 0.173 0.9638 0.504 1.424 0.523 1.477 3.778 0.724 1.606 5.950 0.425 1.575
22 0.640 0.662 0.167 0.9655 0.516 1.415 0.534 1.466 3.819 0.720 1.659 5.979 0.434 1.566
23 0.626 0.647 0.162 0.9670 0.527 1.407 0.545 1.455 3.858 0.716 1.710 5.006 0.443 1.557
24 0.612 0.632 0.157 0.9684 0.538 1.399 0.555 1.445 3.895 0.712 0.759 5.031 0.452 1.548
25 0.600 0.619 0.153 0.9696 0.548 1.392 0.565 1.435 3.895 0.709 1.804 5.058 0.459 1.541

Example 33.1 In a process to manufacture shafts, five observations of shaft diameter of

10 samples are taken on an hourly basis. The recorded observations are given below. Draw X and
1? chart.

Sample No. Sample Value ii (mean) R (Range)

1 8.010.. 8.012 8.011 8.0105 8.0125 8:01i2 0.0025
2 8.000 8.065 8.005 8.002 8.010 8.0164 0.065
3 8.105 8.015 8.008 8.01 8.02 8.0316 0.097
4 8.05 8.06 8.06 8.07 8.00 8.048 0.07
5 8.012 8.015 8.06 8.065 8.012 8.0328 0.053
6 8.015 8.06 8.07 8.07 8.08 8.059 0.065
7 8.06 8.065 8.065 8.105 8.06 8.071 0.045
8 8.03 8.06 8.04 8.06 8.02 8.042 0.04
9 8.04 8.07 8.03 8.04 8.01 8.038 0.06
10 8.06 8.00 8.12 8.04 8.01 8.046 0.12

X = 8.0342 i? = 0.0618
STATISTICAL QUALITY CONTROL 515

The mean of each sample (i.e., T) is calculated by adding all sample value of a row and dividing
by 5 (i.e., number of values in each sample).
The mean of sample mean is calculated by adding the second-last column of and dividing by
10 (i.e., number of times samples are taken).
Range of each sample is calculated in the last column by subtracting the highest value of a sample
from its lowest value. For example: in sample 1, the highest observation is 8.0125 and the least. is 8.01.
Therefore, the range for this sample is (8.0125-8.01) or, 0.0025.
Average-of-range (R) is calculated by adding the last column and dividing by 10.
Calculation of Control Limit: For n = 5; D4 = 2.115, D3 = 0, A, = 0.577 (refer Table 33.3).
For x chart
Upper Control Limit (UCL) = :f: + A, T? = 8.0342 + 0.577x 0.0618 = 8.0698
Lower Control Limit (UCL) = x - A2 T? = 8.0342 - 0.577 x 0.618 = 7.9985
For R Chart
UCL = D4 R = 2.115 x'0.0618 = 0.1307
LCL = D3 R = 0
How to plot .)7 and R chart?
On a graph paper draw UCL, LCL and r lines as shown in the first Figure 33.2, for 7-chart.
Similarly, just below the 7-chart, draw R chart. For this, first draw lines for R , UCL and LCL.
For each sample, plot the values of x on the a -chart and R value on R chart.

8.0742 - 11( .1. = 8.0698

8.0642 -
8.0542 -
8.0442 -
8.0342
8.0242 - = 8.0342

8.0142 -
8.0042 -
7.9942 - = 7.9985

I 2 3 4 5 6 7 9 I0
Sample Number
0.14 - UCI.= 0.1307 .
r i
0.12 -
0.1 -
0.08 -
R =0.6.18
0.06
0.04 -
0.02 -
0 IX' . 0
2 3 4 5 6 7 8 9 10
Sample Number
Figure 33.2 k and R Charts for given Example 33.1
516 INDUSTRIAL ENGINEERING AND MANAGEMENT

Some Preliminary Observations regarding Process

1. The process is out-of-control for 7th sample as the value (8.071) goes beyond VCL. This means
that some preventive action might have been taken at this stage to bring the process in
control.
2. In the beginning of the .7 chart, process is below x line but later, i.e., after 3rd sample, its tendency
is to be above x line. This requires attention of the operator.
3. Range chart behaviour is normal except the, last sample. Since the tendency is to go up, there
is a need to look into the process so that variability in the process is reduced.

Table 33.4 Attribute Control Chart •

Features p-chart np-chart C-chart U-Chart

Purpose For proportion Number of units Number of non- Number of non-
of units non- non-conforming conformities or conformities per
conforming (from (from. sample of defects per unit unit (for samples
sample not constant size) (for sample of not necessarily of
necessarily of constant size) constant size)
constant size)

UCL - 34/5(I - f/) E+

3,fc
": 3 it
P rf +
rr 11

Control Limit

- 3 41
3 0 -13) 3
LCL-4 P +
,4; qn

Central Line p = E p,/n nT, = npilit E= =

Use Result of accept/ Number of For total count When the oppor-
reject type of rejected items of non-conformities tunities for occurrence
inspection in sub-group of in an article (follows of non-conformity
constant size Poisson's distribution) change from subgroup
to subgroup

Example 33.2 Data for the test results of 17 samples of 300 items 'are shown below Draw
p-chart.
Sample number I 2 3 4 5 6 7 8 9 /0 II 12 /3 14 15 16 17
Number of 30 36 33 27 33 39 24 24 12 39 /8 27 24 33 ?I 30 6
defectives
Solution: For each sample, the fraction defective is calculated by dividing the corresponding number
of defectives by total number of items in each run (which is equal to 300).
3VT, (1- 15)
The control limit = P±
4T,
Here, =
EnP 456
= 0.089
n 17 x 300
STATISTICAL QUALITY CONTROL 517

Sample number Number of defectives Proportion fraction defective

I 30 0.10
2 36 0.12
3 33 0.11
4 27 0.09
5 33 0.11
6 39 0.13
7 24 0.08
8 24 0.08
9 12 0.04
10 39 0.13
11 18 0.06
12 27 0.09
13 24 0.08
14 33 0.11
15 21 0.07
16 30 0.1
17 6 0.02
Total 456 11.52

Note that a
means here as: total number inspected during the period.
(1 - 0.089)
UCL = 0.089 + 3.10.089 7 = 0.089 + 0.207 = 0.296
1
\I0.089 (1- 0.089)
LCL = 0.089 + 3 = 0.089 - 0.207 = -0.118 = 0
7
Lower control limit has come out as negative, which is unacceptable; hence it is taken as zero.
ExOmple 33.3 Draw np chart for the data of defectives in Example 33.2.

Solution:
E np
=
k
where, k = number of sample
E np = total number of defectives in k samples
46
np = 26.8
1'7
Control Limits for yip chart
UCL = np + (1 - 75)
LCL = np - 3.\k, (1 -
2
Here, p = - =- 6.8=0.089
n 300
Thus, UCL = 26.8 + 3V26.8 (1- 0.089) = 41.6
LCL = 26.8 - 3V26.8 (1- 0.089) = 11.97
518 INDUSTRIAL ENGINEERING AND MANAGEMENT

-n
UCL, =O.29(

ro
o z
UU
U
IC rE

-I- 00

— N-

-'0

_ r

0
.f) fl .41
O Cl 6 0 0
6 6 O • SOA9▪ 3313(1 JO
S3A!133,10a U0!1.1d0.1(1

Figure 33.3 p and np Chart for Examples 33.2 and 33.3

STATISTICAL QUALITY CONTROL 519

Upper Control Limit

Central Line

Lower Control Limit

Normal Behavior One plot out above. One plot out below. Invest-
Investigate for cause of igate for cause of
poor performance improvement.
Upper Control Limit

Central Line

Lower Control Limit

Two plot near upper control. Two plot near lower control. Run of 5 above central line
Investigate for cause of poor Investigate for cause of poor Investigate l'or cause of
perlbrmance. • improvement. sustained poor performance.
Upper Control Limit

Central Line

Lower Control Limit

Run of 5 below central line. Trened in either direction 5 Erratiee behaviour.
Investigate for cause of plots. Investigate for cause Investigate.
sustained improvement. of progressive change.
Upper Control Limit

Central Line

Lower Control Limit

Sudden change in level.
Investigate for cause.

Time "lime I I- Time

Figure 33.4 Control Chart Evidences f6r Investigation

33.4 ACCEPTANCE 'PLAN

Acceptance plan or acceptance sampling is related to the aspect of quality assurance through inspection.
The basic aim in TQM is to derive customer satisfaction through goad quality product and services.
To check the quality, it is not economical to inspect all the products, coming out of the processes. Therefore,
out of a bigger lot, a small sample is checked for defectives, and sampling plan is designed to achieve
certain degree of quality assurance. •

Acceptance plan is the overall scheme for the acceptance or rejection of a lot based on information
gained from the sample. It is used to identify both the size and type of sample and the criteria
to be used to either accept or reject the lot.
 520 INDUSTRIAL ENGINEERING AND MANAGEMENT

The sampling in the acceptance plan may be single, double or sequential. Based on these, different
acceptance plans are designed:

33.4.1 Advantages of Acceptance Plan

• It is an economical way to ensure good quality production. In general, 100 percent inspection is
too costly. For high volume transfer-line and continuous production process, it is practically impossible
to have 100% inspection.

• • Easy to understand and conduct the plan for quality assurance.

• Sampling is effective in reduced fatigue and cost associated with inspection.
• Easy to train the personnel involved in inspection due to repetitive job.
• Well-documented literature is available for using different plans and strategies related to quality
assurance.
• Easy computation.
• Smaller inspection staff is needed.
• Lesser damage of products is likely due to fewer items being subjected to handling during inspection.

33.4.2 Disadvantages of Acceptance Plan

• The approach is based on sampling; hence no definite statement regarding outgoing quality is possible.
The decision related to acceptance/rejection of the lot is dependent on sampling, which gives only
a probabilistic estimate whether an outgoing lot is defective or not.

33.4.3 Suitability
• For mass production unit, where repetitive jobs of same type come out.
• Where testing is done after breaking or destroying the. work-piece.
This testing is also called as destructive testing such as one used in stress testing, fatigue testing,
etc.
• When inspection cost is very high, as sampling would save number of jobs to be inspected.

33.5 ACCEPTANCE SAMPLING

Inspection is an important way to assure quality. It is generally convenient to inspect items in batches.
Acceptance sampling provides an established way to use statistics in inspection for ensuring quality. It
involves the application of statistical sampling plans, which prescribe the methodology for selecting sample
from a finished output and analysing the information of the sampling plan in maintaining a desired level
of quality.

Use of Acceptance Sampling

• Monitor the quality of incoming material or parts at a desired level.
• Monitor an inspection plan, which involves taking an accept/reject decision about the quality
of big batch of items.
• Monitor the after-effect of the process, i.e., only after the production ends (rather than
process capabilities tests like 7 , R charts) which are process-control techniques.
STATISTICAL QUALITY CONTROL 521

Let us define the following symbols:

N: Number of items in the batch
n: Number of items selected as sample in the N
C: Acceptance number, i.e., in a sample of n, if more than C defectives are found after inspection,
the entire lot is rejected.
AQL: Acceptable Quality L :1
LTPT: Lot Tolerance Percent Defective
a : Producer's risk
{3 Consumer's risk
We will explain these terms in a single sampling plan. Single sampling plan is always defined
by two terms: sample size (n) and acceptance number (c). Thus, if the inspection of n items (out of
a lot of size N) results in more than c defectives, the entire lot is rejected.
How to determine the sample size and acceptance number? To answer this, let us understand the
process and objectives of sampling.

33.5.1 Nomenclature and Symbols used in Sampling Plan

We define the following nomenclatures in a sampling plan:
N: Number of items in a batch (also called as lot size)
N: Number of items selected as a sample in a bigger lot of size N (also called as sample size)
C: Acceptance number, i.e., in a sample of it if more than c defectives are found after inspection,
the entire lot is rejected.
M: Number of defective pieces in a given sample of size it.
M: Number of defective pieces in a given lot of size N.
p: Fraction defective:
In a sample; p =
In a lot; p = MIN.
: Actual process average fraction defective of a product due for inspection
P: Average fraction defectives in a sample
pa: Probability of acceptance of a lot
a : Producer's risk, defined as: probability of a batch being good or even better than AQL (Acceptable
Quality Level) but yielding a bad sample and thus getting rejected. Thus, it is the probability of
rejecting a good lot, which otherwise would have been accepted. •
: Consumer's risk, defined as: probability of a lot being bad or even worse than the limiting quality
but yielding a good sample and thus getting accepted. Thus, it is the probability of a defective
batch being accepted which otherwise would have been rejected.
AQL: Acceptable quality level, defined as the proportion of had components in a lot such that the
lots having less than this proportion have high probability of getting accepted.
LTPD: Lot Tolerance Percentage Defectivs, defined as the proportion of bad components (somewhat
larger than AQL) in a lot such that the lot having more than this proportion of defective components
have a small probability of getting accepted.
In the context of sampling plan, the ideal objective of passing a good lot (with, percentage defectives
less than acceptable quality level) and rejecting a bad lot (with percentage defectives more than
522 INDUSTRIAL ENGINEERING AND MANAGEMENT

AQL) may not be achieved. This is due to two types of sampling errors: type I error and type
II error.
Type I Error: A good lot is rejected due to excess capturing of defectives in the sample. This error
is due to chance factor associated in sampling. However, if this happens, the producer is at a loss
or risk. Hence, type I error is measured by producer's risk.
Type II Error: A bad lot is accepted due to less capturing of defective in the sample. This error is
due to chance factor associated in sampling. However, if this happens, the consumer is at risk
or loss. Hence, type II error is measured by consumer 's risk.

33.6 SINGLE SAMPLING PLAN (SSP)

In this, the decision regarding the acceptance or rejection is made after drawing a sample from a bigger
lot. Inspection is done and if the defectives exceed a certain number the lot is rejected. Otherwise, the
lot is accepted when the number of defectives are less than the acceptance number. A flow chart for
SSP is presented below in Figure 33.5.

Select a Lot of N
Items

Select Random
----'
Remaining (N—n) Items
Sample of
n Items

Defectives found Remove defectives

in sample = C
Inspect all
n Items
If
C<C Replace defectives

Reject
Lot

n non-defective parts

Accept the lot

Figure 33.5 Flow Chart for Single Sampling Plan

33.7 DOUBLE SAMPLING PLAN (DSP)

In this, a small sample is first drawn. If the number of Ifectives is less than or equal to an acceptance
l
number (C1 ), the lot is accepted. If the number of defectives is more than another acceptance number
(C2), which is higher than CI' then the lot is rejected. If in case, the number of defectives in the inspection
STATISTICAL QUALITY CONTROL 523

lies between C2 and C1 , then a second sample is drawn. The entire lot is accepted or rejected on the
basis of outcome of second inspection.
A flow chart of DSP is given in Figure 33.6.

Select a Lot of N Items Select Random

Sample of
Items}
./

Remaining (N-n) Items

C1' Defective
found in this sample Replace defectives
n t Non-defectives
Accept
Lot

(N — n1) Items
Random Sample
()In, Items

Reject
Lot
Inspect
C; Defectives Replace defectives
Found in this sample
112 non-defectives

Figure 33.6 Flow Chart of Double Sampling Plan

33.8 SEQUENTIAL SAMPLING PLAN

In this, items are selected randomly from the lot (N) and tested one-by-one. After each individual testing,
these possible decisions exist: (i) accept, (ii) reject or, (iii) continue sampling. The continue sampling
case is continued till the lot is either accepted or rejected.
For example, in Figure 33.7, in case I, the lot gets one defective after . 15th sample, second defective
after 25th sample, third defective after 30th sample. Till then it is in the continue sample zone, and
thus no decision regarding acceptance/rejection is taken. The fourth defective occurs at 40th sample,
524 INDUSTRIAL ENGINEERING AND MANAGEMENT

and this brings to a decision for the rejection of the lot. In case 2, till 100th sample, only 3 defectives
occurred in the continue zone of sampling. After this, the plan states that accept the lot (and stop
the sampling). Thus, decision regarding acceptance/rejection of the lot is taken on sequential
basis.
Number of Defectives

20 30 40 50 60 70 80 90 100 110 120 130 140 150

Units Sampled (n)

Figure 33.7 Sequential-Sampling Plans

33.9 THE OPERATING CHARACTERISTICS CURVE

The operating characteristics curve (0C-curve) for a quality attribute sampling plan is a graphical
representation of fraction defectives in a lot against the probability of acceptance. It allows us to compare
sampling plans so as to understand the behaviour of different lots with unknown, varying fraction
defectives.

The purpose of 0C-curve is to understand how well a particular sampling plan is effective
in discriminating between a good lot and a bad lot.

What should be the ideal sampling plan? The answer is: an ideal sampling plan should accept
(or pass) a sample, which is drawn from a lot with no defectives, otherwise (i.e., if there is any defective)
the plan should reject the sample. This means that at and above the acceptable quality level, AQL, the
acceptance of lot should be 100 percent of the time. Similarly, with a quality level worse than AQL,
STATISTICAL QUALITY CONTROL 525

•
there should be zero percent chance of accepting the lot. How can this happen? This can happen when
there is 100% inspection. Theoretically, this is no sampling. Hence, we call it an ideal situation and
the OC curve for this case is called as ideal OC-curve. However, the sampling plan should be such
that it should resemble more like ideal 0C-curve, which is very-very steep (in fact, ideal 0C-curve is
vertical at AQL), as shown by dashed lines in Figure 33.8.

Probability of
Accepting Lots
(Percent)

I
100
Producer's Risk (a)

90

.1
80 -

70 -

60 -

pa 50 -
T I
so
40 - n=N

(Ideal OC-Curve)
30

Actual 0C-Curve
20

Objectionable quality region

10
Indifferent Quality region li
r
2 3 4 5 6 7 8 9 10 11 * 12

1' Percent Defective in Lot

(AQL) (UPD)

Figure 33.8 OC Curve for Different Sample Sizes

33.9.1 How to Draw the OC-Curve?

The result of inspection of all the items in a single sampling follows the bionomial distribution as the
outcome of the inspection is binary: either accept or reject the lot. The probability of accepting a lot
(of size N) is equal to the probability of taking a sample of size n with proportion defectives p, and
finding c or fewer defectives items after inspecting the sample. Practically, if sample size (n) is more
than 20 and proportion defective (p) lesser than 0.05, Poisson's distribution may be approximated for
bionomial distribution. Therefore, for such situations, we use Poisson's table (Appendix 33.A) for drawing
OC curves.
526 INDUSTRIAL ENGINEERING AND MANAGEMENT

33.9.2 Procedure of Drawing OC-Curve

Step 1: Find N, n and c.
Step 2: Find the probability of accepting the lot for different values of proportion defectives (i.e.,
p). For this, do the followings:
(a) multiply sample size and proportion defectives, i.e., calculate (np) for each value
of p.
(b) Use Poisson's Table (Appendix 33.A). For each value of np find the column for C.
(c) Note down the value of probability of acceptance (Pa ) for each set of p.
Step 3: Plot Pa on Y axis and p on - X axis.
Step 4: Find a and 13 at AQL and LTPD
at p = AQL; a (producer's risk) is (1 — Pa )
at p = LTPD; p (consumer's risk) is Pa.
Example 33.4 For a lot of 4000 items, a sample of size 80 is drawn each time. The acceptance
number is MO for a single-sampling plan. Draw 0C-curve.
Find producer's risk and consumer's risk at an AQL and LTPD I and 6 percentage defectives
per lot respectively.
Solution: Given:
N: Lot size = 4000
in Sample size = 80
C: Acceptance number
=2
AQL: Acceptable Quality Level
• = 1%
LTPD: Lot Tolerance percentage defectives
= 6%
Calculation of Probability of acceptance (Pa)

Percentage Proportion np Pa = [P (x < 2)] Remark

defectives defectives (p) = 80p Probability of accepting
defectives less than 2 (with
reference to table of Poisson's
distribution: Appendix 33.A)

1 (AQL) 0.01 0.8 0.953 a = 1 — 0.953 = 047

2 0.02 1.6 0.783
3 0.03 2.4 0.570
4 0.04 3.2 0.380
5 0.05 4.0 0.238
6 (LTPD) 0.06 4.8 0.143 = 0.143
7 0.07 5.6 0.082
8 0.08 6.4 0.046
STATISTICAL QUALITY CONTROL 527

Probability of
Accepting lot (P.)

1.00 i Producer's Risk (a) = I 0.953 = 0.047

0.953

.80 N = 4.000 LTPD = 6%

n = 80 AQL= I%
c=2
ar .60

1
.40

.20

0.046
.00
2 .3 4 5 6 7 8
Percentage defective in Lot

Figure 33.9 OC-Curve for example 33.4

33.9.3 Different Conditions in the DC-Curve

Management is always concerned about the selection of sample size (n) and acceptance number (c).
This worry is mainly for:
• Select an 0C-curve, which can better discriminate between a good lot and a bad lot.
For this, the aim of an 0C-curve is:
• To select a plan, which reduces the probability of rejecting good lot and accepting a bad lot. In
other words, it is fair to both producer (as being able to say that a good produced lot is really
termed as good after accepting it) and consumer (as being able to say that a bad lot is really
termed as bad lot after rejecting it).
Effect of Sample Size on 0C-curve: For a given acceptance number (c), us the sample size increases,
the 0C-curve becomes steeper, i.e., more like a vertical line. This increase in n causes increased value
of producer's risk and reduced value of consumer's risk.
For example, in Example 33.4, at AQL (= 1%) and LTPD (= 6%), the value of a and 13 at different
value of n are:

AQL = 1%; LTPD = 6%; C = 2

n a = ( 1—Pa) at AQL 13 = P. at LTPD

80 I — 0.953 = 0.047 0.143

100 I — 0.920 = 0.080 0.062
120 1 — 0.879 = 0.121 0.025
. It is clear that the producer's risk increases and consumer's risk decreases as sample size increases
at a particular value of c. This does what? The higher sample size makes the acceptance of consumer
tougher. Why? In the above table, as lot size increases from 80 to 120, the probability of having two
528 INDUSTRIAL ENGINEERING AND MANAGEMENT

or more defectives (as, c = 2) is more and, thus, likelihood of finding two or more defectives is more.
This means consumers are at lesser risk to receive a bad lot. Similarly, more chances of finding defectives,
"equal to or more than c" with increasing value of n, causes producer a risk due to higher chances
of getting the lot rejected.
The effect of increasing value of n (from 60 to 120) at a constant level of acceptance number
(c = 1) is demonstrated in Figure 33.10. The 0C-curve becomes steeper; hence more discriminating
with increase in n.

1.0
Probability ofAccepting Lots (Percent)

0.9

0.8 n 60, c = 1

0.7 Acceptance Numbers = 1

n = 80, c.= 1 = Constant
0.6

0.5
0.4 n = I 00, r = 1

0.3

0.2 n = 120, c = 1

0.1
0.0
I 2 3 4 5 6 7 8 9 10
(AQ1.) (1.TPD)
Proportion defective (percentage)

Figure 33.10 Effect of Increase in Sample Size (n) on 0C-curve

Effect of Acceptance Level on 0C-curve: As acceptance number (c) increases, 0C-curve becomes
flatter at a particular value of n. This makes the curve less discriminatory. Let us examine the
Example 33.4 at different value of c:

AOL = 1%; LTPD = 6%; n = 80

np at AQL = 0_01 x 80 = 0.8; np at LTPD = 0.06 x 80 = 4.8

C Producer's Risk a Consumer's Risk p

= (1—Pa) at np = 0.8 =Pa at np = 4.8

1 1 — 0.809 = 0.191 0.048

2 I — 0.953 = 0.047 0.143
3 1 — 0.991 = 0.009 0.294
4 I — 0.999 = 0.001 0.476

Thus, -the value of producer's risk decreases and consumer's risk increases. This means a non-
discriminatory sample plan with increasing value of c.
The effect of increasing value of acceptance number, c (from 1 to 4), at a constant level of sainpk
size (n = 60) is demonstrated in Figure 33.11. The 0C-curve becomes flatter; hence less discriminating
with increase in acceptance number.
STATISTICAL QUALITY CONTROL 529

Sample Size = 60
1.0 = constant
\ n=60,c=1
0.9 = 60, c = 2
41 ii= 60,c=3
Probability of Accepta nce

0.8
0.7 4 n = 60, c = 4

0.6
110por
0.5
0.4
0.3
0.2
0.1

I 2 3 4 5 6 7 8 9 10 •
AQL (LTPD)
Proportion Defective (Percentage)

Figure 33.11 Effect of Change in Acceptance Number (c)

on 0C-curve

33.10 AVERAGE OUTGOING QUALITY (AOQ)

In a single sampling plan, the samples are inspected for quality. If a defective is found, it is replaced
with good items. If the total number of defectives exceeds the acceptance number, the lot is rejected
and 100% inspection is done to weed out the defectives. This approach is also termed as rectified inspection.
The acceptance sampling automatically adjusts the level of inspection to the quality of lots being inspected,
as rejected lots are subjected to 100% inspection. To check the performance of the plan, we would like
to check the plan's average outgoing quality (AOQ):

P(P )(N — n)
AOQ
N

where, p = Fraction defectives .

Pa = Probability of acceptance
N = Lot size •
n = Sample size.
We will plot the curve for Example 33.4.
Example 33.5 Plot the curve for average outgoing quality for Example 33.4, when N = 4000,
n = 80, C = 2, LTPD = 6%, AQL = 1%.
Solution: We have already calculated pa for each value of p. Now, using formula for AOQ:
N —n 4000 — 80
AOQ = (Pa )(P) =(13,)(11)
N 4000
= (pa )(p)
530 INDUSTRIAL ENGINEERING AND MANAGEMENT

Incoming defective p = f/100 Pa from AOQ = pp, AOQ in %

fraction (f) in % Example 33.4

1 0.01 0.953 0.0095 0.95

2 0.02 0.783 0.0156 1.56
3 0.03 0.570 0.0171 1.71
4 0.04 0.380 0.0152 1.52
5 0.05 0.238 0.0119 1.19
6 0.06 0.143 0.00858 0.858
7 0.07 0.082 0.00574 0.574
8 0.08 0.046 0.00368 0.368

We plot p and AOQ in Figure 33,12. It shows the performance of plan over a range of possible
proportion of defectives in order to judge whether the plan provides an acceptable degree of protection.
The peak of AOQ curve gives Average Outgoing Quality Limit (AOQL).

1.8

1.6

1 1.4

1.2

1.0

0 V 0.8

(4. ,E,d5
c = 0.6
1
gP
0.4

0.2

0 2 3 4 5 6 7 8

Incoming Fraction Defective (1)

Figure 33.12 AOQ Curve

33.10.1 A Special Feature of AOQ Curve

As fraction defective increases from zero onwards,.there comes a point at which the outgoing fraction
defective (i.e., AOQ) starts beginning to improve. This is due to the fact that sampling plan rejects most
bad lots and they are rectified (i.e., defectives replaced by good items) through 100% inspection. Thus,
STATISTICAL QUALITY CONTROL 531

the level of inspection automatically adjusts to the quality of lots being inspected, assuming rejected
lots are subjected to 100% inspection.
Secondly, AOQL gives an indication that no matter what the incoming fraction defectives are present
in lot, the long-rim average of outgoing fraction defectives will never be worse than AOQL.
AOQL can also be Computed by the formula and table below: •

where; Y = AOQL factor;

n = Sample Size
N = Lot size.

Acceptance 0. 1 2 3 4 5 6 7 8 9 10
Number (C)
AOQL 0.3679 0.84 1.3711 1.9424 2.5435 3.1682 3.812 4.472 5.1457 5.8314 6.5277
Factor (Y)
For, example, in the previous problem, when c = 2, n = 80, N = 4000;
y is 1.3711 (for c = 2 in above table)
Thus, AOQL = 1.3711 (1/80 — 1/4000) = 0.017

The X-bar and s-charts

The main advantage of using the range as a measure of within sample variability is the fact that it
is very easy to calculate. However, assuming the existence of some computing capability, the use of
standard deviation is a preferable alternative. The procedure is exactly the same as for the case of
the X-bar and R-chart except that different coefficients are used for the calculation of the control limits
(Table 33:5).

'rible 33.5 Coefficients for variables charts (X- „and s-charts)

Sample Control Limit Coefficients Divisors for

size Average chart s-chart estimation of a
n A3 C4
83 B4
2 2.659 0 3.267 0.7979
3 L954 . 0 2.568 ,0.8862
4 1.628 0 2.266 0.9213
5 1.427 0 2.089 0.9400
6 1.287 0.030 1.970 0.9515
7 1.182 0.118 1.882 0.9594
8 1.099 0.185 1.815 0.9650
9 1.032 0.239 1.761 0.9693
10 0.975 0.284 1.71-6 0.9727
(Contd..)
532 INDUSTRIAL ENGINEERING AND MANAGEMENT

Sample Control Limit Coefficients Divisors for

size Average chart s-chart estimation of a
n A3 C4
B3 B4

1I 0.927 0.321 1.679 0.9754

12 0.886 0.354 1.646 0.9776
13 0.850 0.382 1.618 0.9794
14 0.817 0.406 1.594 0.9810
15 0.789 0.428 1.572 0.9823
16 0.763 0.448 1.552 0.9835
17 0.739 0.466 1.534 0.9845
18 0.718 0.482 1.518 0.9854
19 0.698 0.497 1.503 0.9862
20 0.680 0.510 1.490 0.9869
21 0.663 0.523 1.477 0.9876
22 0.647 0.534 1.466 0.9882
23 0.633 0.545 1.455 0.9887
24 0.619 0.555 1.445 0.9892
25 0.606 0.565 1.435 0.9896
We calculate the deviation using the formula given below:

(X - X)2
s=
n -.1
As in the case of R, a central value for s is needed. This is obtained by the following formula:

=
Es
k
where, k is the number of samples.
The control limits for the X-chart are given by:
UCL (I) = X + A3T
LCL = X; - A3T
whereas the control limits for the s-chart are given by:
UCL(s) = B4 T
LCL(s) = B3
where, the coefficients A3, B3 and B4 depend on the sample size n and can be found in Table 33.3. For
example, if the sample size is 5, the values are A3 = 1.427, B3 = 0, B4 = 2.089.
u-chart: The u-chart is another kind of attribute chart and is used for process control for defects
when it is not possible to take a sample of constant sample size. As in the case of the p-chart, the
data values plotted on the chart are the proportions of faults per sample, now symbolized by u.
STATISTICAL QUALITY CONTROL 533

Note that it is the number of non-conformities per item within the sample that is being monitored,
and not the number of items per sample which have been rejected; a p-chart would have been appropriate
if the number of items per sample which have been rejected is monitored.
The control limits are given by:
117—
/7 + 3

where ii is the average of all sample sizes and i i is calculated as:

Total number of defects
=
Total number of items inspected
c-chart: The c-chart is used for process control of defects when it is possible to take samples
at a constant sample size. The data plotted on the chart are the number of defects c in each sample.
The control limits are given by:
±
where, E is the average number of defects calculated by:
_ Total number of defects
c =
total of samples tested
A control chart tells us whether or not a process is in statistical control, or whether or not more
than common causes of variation are present in the process. It informs us to determine and eliminate
any special causes of variation present. The chart characteristics of a process operating with only common
causes of variation are:
1. All points fall within the control limits.
2. An approximately equal number of points on either side of the central line.
3. There is no specific pattern apparent.
4. Most points are near the centre (the middle third region) although a few (about one-third) approach
the control lines.
The chart not satisfying the above characteristic is pinpointing the presence of special causes of
variation, which needs investigation and, if required, elimination of such causes. We can therefore define.
certain action rules applicable to all control charts in general; these rules determine the circumstances
when action should be taken.
An action does not mean an action for elimination (of the special cause); it means an action for
investigation for the cause. For example, a point falls below the lower control limit of an attribute chart
for defects or of. a range chart for variables (if such a limit exists that is positive), should prompt an
investigation of the circumstances for such a situation, so that it can be repeated again in the
future.
Some of the basic action rules are given below:
Rule 1: Action should be taken when at least one of the plotted points lies outside the (3o) control
limits. If the motive for the special cause is understood and permanently taken care of, the out of control
reading can be removed and the new control limits are calculated. If the reason is not found, then it
must be accepted that the sample value is part of the system which generated the limits.
Rule 2: (rule of seven). This calls for action if there are seven consecutive points, either all on
one side of the mean, or all increasing, or all decreasing. Since the probability of any of these alternatives
occurring by chance is very small, a special cause in need of investigation might be present.
534 INDUSTRIAL ENGINEERING AND MANAGEMENT

Rule 3: Action should be taken when non-random patterns are clearly shown in the chart. A cyclic
pattern might indicate that the material from a specific supplier, or the circumstances of a specific time
period, might regularly have an undesired (or beneficial) effect.
Rule 4: (Middle third rule). Action should be taken when the number of plotted points in the
middle third of the total range between the control lines is much less than two-thirds of the total number
of points. Care should be taken when applying this rule in situations when the distribution of the readings
is skewed, as in the range chart.
Limits of warning: Most of the SPC practitioners prefer the use ofthe warning limits; these are
lines drawn at ± 2a from the mean, in addition to the control lines. (at ± 3a from the mean) which
are now called action limits. The warning lines can easily be drawn at two-thirds of the distance (outwards)
between the central line and each control limit. Care should be .taken in the case of non-symmetrical
distributions, as in the range charts.
SPC in Service Sector
The service sector is growing in a big way now-a-days. The Statistical Process Control is applicable
. to both manufacturing and service areas. There is a need to establish an appropriate characteristic which
is tangible. Obviously, this is much easier for the manufacturing sector, where the production process
is much more clearly quantifiable. However, 80 per cent of activities in manufacturing sector is service
like personnel, administration, procurement, materials handling, design etc. It is important, therefore, to
be able to define measurable characteristics, appropriate to the service area which, at the end of the
day, is equally applicable to the manufacturing area.
Administration
• Average number of typing errors per document.
• Typing throughput time
• Proportion of reports delivered to schedule.
• Customer complaint rate.
• Number of debtors or creditors outstanding.
• Number of projects/customer orders completed 'to schedule.'
• Number of telephone rings before answering.
• Improperly directed calls to switchboard.
• Number of customers waiting on line.
• Processing errors.
• Downtime in processing equipment.
• Time spent to locate documents.
• Timelines of courier deliveries.
• Payment errors.
Personnel
• Recruitment rate.
• Turnover rate.
• Cost per recruitment.
• Time taken to fill the vacancies.
• Number of resignations.
• Unfulfilled planned training.
STATISTICAL QUALITY CONTROL 535

• Attrition rate
• Absenteeism.
Procurement
• Proportion of purchasing orders completed on time.
• Average order throughput time.
• Average time spent chasing orders.
• Supplier quality performance.
• Number of return to supplier items.
• Number of purchasing errors due to the purchaser.
Materials handling
• Average time of out of stock periods.
• Ratio of stock-out to demand.
• Material throughput time.
• Proportion of internal orders concluded on time.
• Number of stock items not used as planned.
• Response time to customer requests.
Design
• Number of drawing errors.
• Number of checking mistakes.
• Number of void designs.
• Proportion of. drawings not completed on schedule.
• Timeliness of design-error resolution.
• Missing or illegible drawings.