Chapter 8 Statistical Process Control

8 STATISTICAL
PROCESS
CONTROL
Objectives
After studying this chapter you should

understand the purpose of statistical process control;

be able to set up and use charts for means, ranges, standard

deviations and proportion non-conforming;

be able to use an appropriate method to estimate short-term

standard deviation;

be able to interpret the variability of a process in relation to

the required tolerances.

8.0

Introduction

Statistical process control may be used when a large number of

similar items - such as Mars bars, jars of jam or car doors are
being produced. Every process is subject to variability. It is not
possible to put exactly the same amount of jam in each jar or to
make every car door of exactly the same width. The variability
present when a process is running well is called the short term
or inherent variability. It is usually measured by the standard
deviation.
Most processes will have a target value. Too much jam in a jar
will be uneconomic to the manufacturer but too little will lead to
customer complaints. A car door which is too wide or too
narrow will not close smoothly.
The purpose of statistical process control is to give a signal
when the process mean has moved away from the target. A
second purpose is to give a signal when item to item variability
has increased. In either case appropriate action must then be
taken by a machine operator or an engineer. Statistics can only
give the signal, the action relies on other skills.

147

Chapter 8 Statistical Process Control

8.1

Control charts

The most common method of statistical process control is to take

samples at regular intervals and to plot the sample mean on a
control chart.
Sample
mean

(3)
(1)

Action limit
Warning limit
Target value

(2)

If the sample mean lies within the warning limits (as point (1)) the
process is assumed to be on target. If it lies outside the action
limits (as point (2) the process is off target and the machine must
be reset or other action taken. If the mean is between the warning
and action limits (as point (3)) this is a signal that the process may
be off target. In this case another sample is taken immediately. If
the mean is still outside the warning limits action is taken. If
however the second sample mean is within the warning limits
production is assumed to be on target.

Setting the limits

If the limits are too far from the target value small deviations from
the target may go undetected but if the limits are too close to the
target value there will be a large number of false alarms (that is
there will be a signal for action when the process mean is on target
and no action is necessary). To decide where to set the limits a
measure of short term variability is needed.

Activity 1
Here are the weights, in grams, of 100 Mars Bars (unwrapped)
59.55
64.57
63.34
58.55
59.80
61.82
59.90
62.17
68.11
60.46

62.33
60.14
60.01
65.36
61.45
58.98
57.73
61.29
65.46
59.05

63.68
62.51
59.11
63.03
60.78
62.63
67.08
69.01
57.81
61.06

67.10
62.02
62.57
61.71
61.89
59.68
63.25
63.31
64.73
55.08

56.85
60.16
58.48
62.26
63.91
62.79
64.20
62.92
63.27
61.60

57.84
61.45
60.25
62.05
58.53
63.90
61.16
64.13
64.63
63.85

64.40
58.42
61.42
60.42
59.29
62.64
61.03
62.46
59.70
64.42

Check that the standard deviation is 2.60 g.

148

60.26
58.19
63.25
58.77
62.24
61.96
65.79
60.61
54.59
62.91

62.05
65.65
63.46
62.69
61.12
64.14
62.43
61.58
61.83
63.54

64.29
65.90
63.33
66.20
60.60
60.70
62.75
60.71
59.21
60.69

Warning limit
Action limit

Chapter 8 Statistical Process Control

If the target weight for Mars bars is 61.5g and production is to be

controlled by weighing samples of size 5 at regular intervals then
the sample means will have a standard deviation of
2.60
= 1.16 g .
5

It has been found convenient in practice to set the warning limits

so that, if the mean is on target, 95% of sample means will lie
within them. The action limits are set so that 99.8% of sample
means lie within them when the mean is on target. Since we are
dealing with sample means it is reasonable to assume they are
normally distributed. Hence in this case the warning limits will
be set at

61.5 1.96

2.6
5

61.5 2.28

59.22 and 63.78

the action limits will be set at

61.5 3.09

2.6
5

61.5 3.59

57.91 and 65.09

In general the warning limits will be set

1.96

and the action limits at

3.09

where is the target value and is the short term standard

deviation (that is the standard deviation over a short period of
time when the mean is constant and only short term variability is
present).
The standard deviation will in practice only be an estimate of the
true value and 95% and 99.8% are somewhat arbitrary figures.
For these reasons and for simplicity the limits are often set at

2
3
and
n
n

149

Chapter 8 Statistical Process Control

Activity 2
(a) Draw a control chart for means for the Mars bars.
(b) Calculate the sample mean for each of the following ten
samples. Plot the means on the chart.
Sample
1

10

59.55 64.57 63.34 58.55 59.80 61.82 59.90 62.17 68.11 60.46
62.33 60.14 60.01 65.36 61.45 58.98 57.73 61.29 65.46 59.05
63.68 62.51 59.11 63.03 60.78 62.63 66.03 69.01 57.81 65.06
67.10 62.02 62.57 60.72 61.89 59.68 64.25 63.31 64.73 55.08
56.85 60.16 58.48 62.26 63.91 62.70 65.20 62.92 63.27 61.60

(c)Have any signals to stop production been given.

Activity 3
A machine filling packets of breakfast cereal is known to operate
with a standard deviation of 3 g. The target is to put 500 g of
cereal in each packet. Production is to be controlled by taking 4
packets at regular intervals and weighing their contents.
(a) Set up a control chart for means.
(b) What action, if any, would you recommend if the next sample
weighed
(i)

503,497,499,496

(ii) 501,491,492,492
(iii) 502,500,507,505
(iv) 500,501,501,501?
In what way are statistical processes controled by hypothesis
testing ralated?

Target values
In some cases the target value is not known or not attainable. For
example the target for strength of malleable iron castings is 'as
large as possible'. The target value for percentage impurity is
probably zero but it will be recognised that this is unattainable (and
even if it was attainable it would be nonsense to set up charts with
lower limits set at negative values of impurity). In these cases a
large sample should be taken when the process is believed to be
running satisfactorily and the sample mean used as a target value.
150

Chapter 8 Statistical Process Control

For example, tests on the tensile strength of malleable iron castings

at a foundry at a time when production was thought to be
satisfactory gave a mean of 148 and a standard deviation of 2.5
(the units are Newtons m 2 10 9 ).
The process is to be controlled by testing the strength of samples
of size 3 at regular intervals.
Warning limits are

148 1.96

2.5
3

145.2 150.8

Action limits are

148 3.09

2.5
3

143.5 152.5

A sample mean less than 143.5 would lead to action being taken in
the usual way. A sample mean greater than 152.5 would indicate
that the average strength had increased. This would not of course
lead to action to reduce the mean but might lead to an investigation
to see how the improvement may be maintained.

Control of variability
Referring back to the Mars bar example, the next sample weighed
59.35 62.46 48.67 68.79 71.23.

The sample mean is 62.1 which is comfortably within the warning

limits and no action is signalled. However closer examination of
this sample shows that something has gone drastically wrong with
production. Although the mean is acceptable the variability within
the sample has increased alarmingly. The means chart is not
reliable for detecting a change in variability. We need an
additional chart for this purpose.
The best measure of variability is the standard deviation.
However, since the sample range is simpler both to understand and
calculate, traditionally it has been used as a measure of variability
in statistical process control. With the ready availability of
calculators there is now little reason for not using the standard
deviation as a measure of variability. Whichever measure is
chosen the charts are constructed in the same way. They depend
on an estimate of standard deviation being available. They also
assume that the data are normally distributed. It is no longer

151

Chapter 8 Statistical Process Control

enough only to assume that the sample means are normally

distributed. No target value is shown on the charts. The target for
variability is zero but this is an unrealistic value to use on the
charts.

8.2

Chart for ranges

The limits for the range are found by multiplying the short term
standard deviation by the appropriate value of D found from tables.
In the case of the Mars bars with samples of size 5 and a short term
standard deviation of 2.6
the upper action limit is

5. 484 2.6 = 14.26

the upper warning limit is

4.197 2.6 = 10.9

the lower warning limit is

0.850 2.6 = 2.21

the lower action limit is

0.367 2.6 = 0.95

Activity 4
Draw a control chart for sample ranges for the Mars bars. Plot the
ranges of the ten samples given in activity 2. Are there any signals
to stop production?

8.3

Charts for sample standard

deviations

These are calculated and operated in exactly the same way as the
charts for ranges. The only difference being that the appropriate
factor is selected from a different table. The standard deviation
chart gives a slightly better chance of detecting an increase in the
variability when one exists. The risk of a false alarm is the same
for both charts.
For the Mars bar example
the upper action limit is

2.15 2.6 = 5.59

the upper warning limit is

1.67 2.6 = 4.34

the lower warning limit is

0.35 2.6 = 0.91

the lower action limit is

152

0.15 2.6 = 0.39

Chapter 8 Statistical Process Control

As with the range charts the lower limits often are not used.
Note the sample standard deviations should be calculated using the

formula =

( x x )2
n 1

Activity 5
Draw a control chart for sample standard deviations for the Mars
bars. Plot the sample standard deviations for the ten samples given
in activity 2. Are there any signals to stop production?

Activity 6
Set up a chart for sample standard deviations for the packets of
breakfast cereal in activity 3.
What action, if any, would you recommend if the next sample
weighed
(i)

502,496,499,496,

(ii) 505,495,496,496,
(iii) 499,490,505,505,
(iv) 506,510,508,508,
(v) 497,498,498,497?
Note that a sample must lie within the warning limits on both the
mean and the standard deviation (or range) chart for production to
be considered satisfactory without further checking.

8.4

Estimating the short term

standard deviation

The best way of estimating the short term standard deviation is to

take a large sample when the process is running well and calculate
the standard deviation using the formula

( x x )2
n 1

This is how the standard deviation was estimated for the Mars
bars. The same sample may also be used for estimating a suitable
target value when one is required. This procedure is called a
process capability study.
153

Chapter 8 Statistical Process Control

Because data is often collected in small samples for statistical

process control purposes the short term standard deviation is often
calculated from a number of small samples. If the process is
running well at the time the standard deviation should be constant
but it may be that there have been some small changes in the mean.
If this is the case pooling the small samples and regarding them as
one large sample will tend to over estimate the short term standard
deviation. It is better to make an estimate of variability from each
sample individually and then take the average. This estimate may
be the sample range or the sample standard deviation.
For example, a company manufactures a drug with a nominal
potency of 5.0 mg cm3. For prescribing purposes it is important
that the mean potency of tablets should be accurate and the
variability low. Ten samples each of four tablets were taken at
regular intervals during a particular day when production was
thought to be satisfactory. The potency of the tablets was
measured.
Sample 1

10

4.97 4.98 5.13 5.03 5.19 5.13 5.16 5.11 5.07 5.11
5.09 5.15 5.05 5.18 5.12 4.96 5.15 5.07 5.11 5.19
5.08 5.08 5.12 5.06 5.10 5.02 4.97 5.09 5.01 5.13
5.06 4.99 5.11 5.05 5.04 5.09 5.09 5.08 4.96 5.17

The ranges of the ten samples are

0.12 0.17 0.08 0.15 0.15 0.17 0.19 0.04 0.15 0.08
giving a mean sample range of 0.13. This mean range can be
converted to a standard deviation by multiplying by an appropriate
factor from the appropriate table. For samples of size 4 the
estimated standard deviation would be
0. 4857 0.13 = 0.063

The factors in the table assume that the data is normally

distributed.
Alternatively the sample standard deviations for the ten samples
are
0.0548
0.0804 0.0359
0.0678 0.0618
0.0753

0.0873

0.0171

0.0660

0.0365

For mathematical reasons, the best way of estimating the standard

deviation is to find the mean of the 10 variances and take the
square root, i.e.
0.05482 + 0.0804 2 + 0.0359 2 + 0.06782 + 0.06182 + 0.07532 + 0.08732 + 0.01712 + 0.0660 2 + 0.03652
10
= 0.062 .

154

Chapter 8 Statistical Process Control

As can be seen, there is little difference in the two estimates.

Although the second method is mathematically preferable the first
method is perfectly adequate for most purposes.

Example 1
In the production of bank notes samples are taken at regular
intervals and a number of measurements made on each note. The
following table shows the width, mm, of the top margin in eight
samples each of size 4. The target value is 9 mm.

Sample 1

9.0 10.4

8.2

7.9

8.2

8.4

7.4

7.6

8.1

9.0

9.2

7.7

9.0

8.1

8.0

8.5

8.7

7.9

7.9

7.7

7.4

8.4

8.9

8.1

7.5

7.2

7.7

9.3

8.6

8.7

9.8

8.8

(a) Calculate the mean sample range and, assuming a normal

distribution, use it to estimate the standard deviation of the
process.
(b) Use the estimate made in (a) to draw a control chart for
means showing 95% warning limits and 99.8% action limits.
Plot the eight means.
(c) Draw a control chart for ranges showing the upper and lower
action and warning limits. Plot the eight ranges.
(d) Comment on the current state of the process.
(e) What action, if any, would you recommend in each of the
following cases.
The next sample is
(i) 9.1, 10.2, 8.9, 9.7
(ii) 7.3, 6.9, 8.8, 7.1
(iii) 10.4, 10.1, 9.2, 6.8
(iv) 10.9, 9.8, 8.8, 11.1
(v) 9.3, 9.2, 9.3, 9.3
(f) Suggest two methods, other than the one used in (a), to
estimate the short term standard deviation of the process.
Compare the relative merits of these three methods in the
context of control charts.
Solution
Sample

mean

8.325

8.625

8.25

8.15

8.3

8.4

8.525

8.25

range

1.5

3.2

1.5

1.6

1.6

0.6

2.4

1.2

155

Chapter 8 Statistical Process Control

13.6
= 1. 7
8
estimated standard deviation = 0. 4857 1. 7 = 0.826

(a) mean range =

(b) chart for means

warning limits 9
action limits 9

1.96 0.826
8.19 9.81
4

3.09 0.826
7. 72 10.28
4

Sample
mean (mm)
Action limit
10

Warning limit
Target value

Warning limit

Action limit

Sample number
1

(c) chart for ranges

upper action limit

5.309 0.826 = 4.39

upper warning limit

3.984 0.826 = 3.29

lower warning limit

0.595 0.826 = 0. 49

lower action limit

0.199 0.826 = 0.16

Sample
Sample
range (mm)
mean (mm)

Action limit
4
Warning limit
3

1
Warning limit
Action limit
Sample number

0
1

156

Chapter 8 Statistical Process Control

(d) All ranges are within warning limits. Variability appears to be

under control. One mean below lower warning limit. This is
not in itself a problem. Since 95% of sample means lie within
the warning limits when the process is on target, 5% or 1 in 20
will lie outside warning limits if mean on target. However as
all sample means are below target it appears that the machine
needs resetting to increase the mean.
(e)

(i)

mean 9.475 range 1.3 no action

(ii) mean 7.525 range 1.9 below action limit on mean

chart. Action needed to increase mean probably
resetting machine.
(iii) mean 9.125 range 3.6 outside warning limit on range
chart. Take another sample immediately. If still outside
warning limit take action to reduce variability. Probably
overhaul machine.
(iv) mean 10.15 range 2.3 Outside warning limit on mean
chart. Take another sample immediately. If still outside
warning limit take action to increase mean.
(v)

mean 9.275 range 0.1 below lower action limit on

range chart. Check data. If correct no action but
investigate how variability has been reduced with a
view to maintaining the improvement.

(f) An alternative method of estimating the standard deviation

is to regard the 32 observations as a single sample and to
estimate the standard deviation using the formula

( x x )2
(n 1) . This would be a satisfactory method if the mean
had not changed while the eight samples were taken but would
tend to overestimate the short term standard deviation if it had.
In this case the eight plotted means suggest that the mean has
remained constant (although below target).
Another alternative is to calculate the standard deviation separately

for each sample. If these values are 1 , 2 , K, 8 the estimate of
the standard deviation will be

12 + 2 2 + K + 8 2
8
another alternative is to use

1 + 2 + K + 8
.
8
This method will be satisfactory even if the mean has changed and
even if the data is not normally distributed.

157

Chapter 8 Statistical Process Control

Additional rules for taking

action
When charts are plotted attention can be focused not only on the
latest point but on the recent history of the process. A large
number of consecutive points above (or below) the centre line
on a chart for means may indicate that the mean is off target
even though no single point has violated the action limits and no
two consecutive points have violated the warning limits.
Similarly an upward or downward trend on a chart may indicate
the necessity for action even though no limits have been
violated.
Rules such as, take action if eight successive points are on the
same side of the target value, or, take action if seven consecutive
points each exceed (or are less than) the previous point have
been used. These additional rules will all increase the ability of
the charts to detect a deviation from the target value but it
should be borne in mind that they will also increase the number
of false alarms.
Why will the number of false alarms be increased?

8.5

Tolerance limits

Many processes have tolerance limits within which the product

must lie. The purpose of statistical process control is to ensure
that a process functions as accurately and with as little
variability as possible. Despite this the process may be unable
to meet the tolerances consistently. Alternatively it may be able
to meet the tolerances easily.
In the example of the drug potency the tolerances were 4.9 to
5.1. The standard deviation was estimated to be about 0.06. It is
generally reasonable to assume that mass produced items will
follow a normal distribution. If we also assume that the mean is
exactly on target, i.e. 5.00, we can calculate the proportion
outside the tolerances.

z1 =

z2 =

( 4.9 5.0 )
0.06

(5.1 5.0 )
0.06

= 1.67

1 0.9524

1.67

= 1.67

The proportion outside the tolerances is 2 (1 0.95254 ) = 0.095 .

Hence even in the best possible case with the mean exactly on
target about 10% of the tablets would be outside the tolerances.

158

1 0.9524

1.67

Chapter 8 Statistical Process Control

This process cannot meet these tolerances however well statistical

process control is applied. A better and almost certainly more
expensive process is needed to meet these tolerances.
Almost all of a normal distribution lies within 3 standard
deviations of the mean. So if the tolerance width exceeds 6
standard deviations the process should be able to meet the
tolerances consistently, provided the mean is kept on target. In the
drug example the tolerance width was 5.1 4.9 = 0.2 which is
0.2
only
= 3.3 standard deviations.
0.06
In some cases the tolerance width may greatly exceed 6 standard
deviations and the tolerances will be easily met. It can be argued
that in these cases statistical process control is unnecessary.
However to produce a high quality product it is better to have the
process exactly on target than just within the tolerances. A car
door which is exactly the right width is likely to close more
smoothly than one which is just within the acceptable tolerances.

Example 2
The copper content of bronze castings has a target value of 80%.
The standard deviation, is known to be 4%. During the production
process, samples of size 6 are taken at regular intervals and their
copper content measured.
(a) Calculate upper and lower warning and action limits for
control charts for
(i) means,
(ii) standard deviations.
(b) The following results were obtained from samples on three
separate occasions
(i) 82.0 83.5 79.8 84.2 80.3 81.0
(ii) 75.8 68.4 80.3 78.2 79.9 73.5
(iii) 79.5 80.0 79.9 79.6 79.9 80.4
For each sample calculate the mean and standard deviation and
recommend any necessary action.
(c) If the process currently has a mean of 76% with a standard
deviation of 4%, what is the probability that the mean of the
next sample will lie within the warning limits?
(d) The tolerance limits are 73% and 87%. A process capability
index, C p , is defined by
(i)

Calculate C p =

tolerance width
6

(ii) Explain why a C p value less than 1 is regarded as

unsatisfactory.
159

Chapter 8 Statistical Process Control

(iii) Explain why, if the mean is off target, a C p value

greater than 1 may still be unsatisfactory.
(e) In addition to using warning and action limits, action is
sometimes recommended if eight consecutive means lie on
the same side of the target value. What is the probability of
the next eight means being on the same side of the target
value if the process mean is exactly on target?
Solution
(a)

Control chart for means

Warning limits 80 1.96

4
6

76.8 83.2

80 3.09

4
6

74.95 85.05

Action limits

Control chart for standard deviations

upper action limit

2.03 4 = 8.12

upper warning limit

1.60 4 = 6. 4

lower warning limit

0. 41 4 = 1.64

lower action limit

0.20 4 = 0.8

(b) (i) mean 81.8 s.d. 1.77 no action

(ii) mean 76.0 s.d. 4.53 mean below warning limit
You should take another sample immediately. If mean still
outside warning limits, take action.
(iii) mean 79.9 s.d. 0.32 s.d. below action limit.
Variability has been reduced; try to find out why so that
improvement may be maintained.
(c) z1 =

z2 =

( 76.8 76)
4

(83.8 76)
4

= 0. 490

= 4. 777

probability within warning limits = 1 0.68793 = 0.312

160

0.68793

0.490

4.777

Chapter 8 Statistical Process Control

(d) (i)

Cp =

(87 73)
= 0.583
(6 4 )

Outside
tolerance

Outside
tolerance

(ii) A C p less than 1 indicates that the tolerance width

is less than 6 . Hence the process is unlikely to be
able to meet the tolerances consistently even if the
mean is exactly on target.
(iii) Even if the C p is less than one the tolerances may

Outside
tolerance

not be consistently met if the mean is off target.

(e) Probability of next eight means being all on the same side of
the target is

(0.5)8 + (0.5)8 0.0078

8.6

Control charts for proportion

non-conforming or defective

Control charts may also be applied when instead of measuring a

variable, such as weight or length, items are classified as
conforming or non-conforming (defective and non-defective used
to be more common terms but are now going out of use).
Samples of 100 components are taken from a production line at
regular intervals and the number non-conforming counted. At a
time when production was thought to be satisfactory 12 samples
contained the following numbers of non-conforming items
10, 13, 12, 19, 8, 14, 17, 16, 10, 18, 9, 16
The total number of non-conforming items found is 162 out of
1200 components examined. This gives an estimate of p, the
162
= 0.135 .
proportion non-conforming, of
1200
If p is constant the number non-conforming in the samples will
follow a binomial distribution. As n is large (100) and np is
reasonably large (100 0.135 = 13.5 ) the binomial distribution may
be approximated by a normal distribution with mean np and
standard deviation

np(1 p) . Control charts may be set up in the

usual way.
In this case warning limits for p are

0.135 1.96 0.135

0.865
100

0.068 0.202

161

Chapter 8 Statistical Process Control

action limits for p are

0.135 3.09 0.135

0.865
100

0.029 0.241

Proportion
non-conforming
0.25
Action limit
Warning limit

0.20

0.15

0.10
Warning limit
0.05
Action limit
Sample number

0
1

10

There are several points to note about this chart

1.

Only one chart is necessary. The only way a signal can be

given is if there are too many non-conforming items.

2. 0.135 is not the target value. The target value is zero.

However in this case 0.135 is thought to be a reasonable level.
The real purpose of these charts is to ensure that the proportion
non-conforming does not rise above 0.135.
3.

The lower limits are not needed but may be retained as a check
on errors in the data or to indicate that an improvement has
occurred.

4.

Charts for the number non-conforming could be plotted as an

alternative to the proportion non-conforming. There may be
an advantage in plotting the proportion if the number of items
inspected is likely to vary slightly from sample to sample.

5.

Any variable such as weight or length may be treated in this

way by defining limits outside which the item is classified as
non-conforming. It is usually easier to decide whether an item
lies within given limits or not rather than to make an exact
measurement. However to get equivalent control of the
process much larger samples are needed.

6.

If the samples are too small the normal approximation will not
be valid. It would be necessary to use exact binomial
probabilities to calculate the limits. However as small samples
do not give good control using proportion defective this
problem is not likely to arise. Small values of p would also
cause the normal approximation to be invalid. However good
control cannot be maintained for small values of p. This
problem can be overcome by tightening up the definition of
non-conforming.

162

11

12

Chapter 8 Statistical Process Control

Example
At a factory making ball bearings, a scoop is used to take a sample
at regular intervals. All the ball bearings in the scoop are
classified as conforming or non-conforming according to whether
or not they fit two gauges. This is a very quick and easy test to
carry out.
Here is the data for ten samples
Sample

Number in scoop 95

99

Number
non-conforming

16

10

115 120

84

107

97

119

92

112

11

11

13

14

10

13

10

(a) Use the data to estimate p, the proportion non-conforming,

and n, the average number in a scoop. Use these estimated
values of n and p to set up control charts for proportion nonconforming. Plot the 10 samples on the chart and comment.
(b) The following samples occurred on separate occasions when
the chart was in operation,
(i)

115 in scoop, 21 non-conforming

(ii) 94 in scoop, 8 non-conforming

(iii) 92 in scoop, 20 non-conforming
(iv) 12 in scoop, 3 non-conforming
(v) 104 in scoop, 1 non-conforming
For each of the samples comment on the current state of the
process and on what action, if any, is necessary.
(AEB)
Solution
(a) The total number non-conforming observed is

16 + 6 + 11 + 10 + 11 + 5 + 13 + 14 + 10 + 13 = 109
the total number of ball bearings observed is
95 + 99 + 115 + 120 + 84 + 107 + 97 + 119 + 92 + 112 = 1040

n=

1040
= 104 ,
10

p=

109
= 0.1048
1040

This gives warning limits

0.1048 1.96 0.1048

0.8952
104

0.046 0.164

163

Chapter 8 Statistical Process Control

and action limits

0.1048 3.09 0.1048

0.8952
104

0.012 0.198

proportion non-conforming in samples

0 .168, 0.061, 0.096, 0 .083, 0.131, 0.047, 0.134, 0.118, 0.109, 0.116
Proportion
non-conforming
0.20

Action limit
Warning limit

0.15

0.10

0.05

Warning limit
Action limit
Sample number

0
1

One point is just outside the warning limits. As one point

in 20 is expected to lie outside the warning limits this is
acceptable and does not cast doubt on the state of the
process when these samples were taken. (If the process was
unstable when the samples were taken they would be
unsuitable for estimating a value of p for use in calculating
limits for the chart. It would be necessary to take new
samples when the process was running well and use these
for estimating p.)
21
= 0.183 above warning limit. Take another
(b) (i)
115
scoop immediately.
8
(ii)
= 0.085 production satisfactory.
94
(iii)

20
= 0.217 above action limit. Take action.
92

(iv) The limits on these charts depend on the sample size

n. A small amount of variability in n will make little
difference but they are unsuitable for use with
samples as small as 12. Take another scoop and start
again.
(v)

164

1
= 0.010 below action limits. Process has
104
improved. Try to maintain the improvement.

10

Chapter 8 Statistical Process Control

8.7

Miscellaneous Exercises

1. A steel maker supplies sheet steel to a

manufacturer whose machines are set on the
assumption that the thickness is 750 (the units
are thousandths of a mm). The supplier decides
to set up control charts to ensure that the steel is
as close to 750 as possible. 8 samples each of
5 measurements are taken when production is
thought to be satisfactory. The sample means
and ranges are as follows:
sample
mean
range

738 762 751 763 757 754 762 761

14 21 17 53 29 49 71 62

(a) Assuming the data are normally distributed

estimate the standard deviation of the
process.
(b) Using the target value as centre line draw a
control chart for means showing 95% and
99.8% control lines. Plot the eight means.

The units are cm above 9 m 50 cm.

Sample
1
2
42
57
36
8

64
76
52
40

10

53
7
41
56

62
31
17
44

26
54
62
51

27
37
69
61

16
54
32
48

17
44
46
49

48
72
76
64

41
59
80
68

(a) Use these samples to calculate limits for

control charts for means and for standard
deviations. Draw the charts and plot the
points.
(b) What action, if any, would you take if the
next sample was
(i) 54, 64, 70, 48,
(ii) 9, 99, 42, 58,
(iii) 84, 106, 99, 93?

(c) Draw a control chart for the range showing

the upper action and warning limits. Plot the
eight ranges.

(c) The tolerance for the length of wire on a

reel is 50 70. Comment on the ability of
the process to meet this.

(d) Comment on the patterns revealed by the

charts and advise the producer how to
proceed.

(d) If the current mean is 55 and the standard

deviation is 18 what is the probability that
the next sample mean will be within

(e) In addition to using warning and action

limits, action is sometimes recommended if
seven consecutive points lie on the same
side of the target value. What is the
probability of this occurring if the process is
exactly on target?
(AEB)
2. Raw material used in a chemical process
contains some impurity.
To ensure that the percentage impurity does not
become too large samples of size 3 are tested at
regular intervals. When the process is running
satisfactorily the mean percentage impurity was
found to be 16.7 with a standard deviation of 3.4.
(a) Set up control charts for means and for
standard deviations.
(b) What action, if any, would you recommend
if the next sample was
(i) 16.9,19.3,20.2
(ii) 24.2,25.6,22.0
(iii) 14.2,19.1,5.2
(iv) 22.7,19.3,23.1
(v) 9.3,12.2,8.1?
3. Reels of wire are wound automatically from a
continuous source of wire. After each reel is
wound the wire is cut and a new reel started.
The target is to wind 10 m on each reel. The
lengths of wire on ten samples each of four reels
are measured with the following results.

(i) the warning limits,

(ii) the action limits?
4. A biscuit factory produces cream crackers.
Packets are sampled at hourly intervals and
the weight, in grammes, of the contents
measured. The results below are from seven
samples each of size five taken at a time when
production was thought to be satisfactory. The
target value is 210 g.
sample
1
2
209.0
211.0
213.5
205.0
210.0

214.5
210.0
212.0
203.5
214.0

204.5
209.0
198.5
203.0
213.5

217.5
216.0
217.0
214.0
209.5

211.0
198.0
209.5
213.5
213.5

224.0
220.0
224.0
218.5
210.0

210.0
208.0
210.0
220.0
211.0

(a) Estimate the standard deviation of the

process.
(b) Calculate upper and lower warning and
action limits for a control chart for
(i) means,
(ii) standard deviations.
(c) Draw the two control charts and plot the
points. Comment on the current state of
the process.

165

Chapter 8 Statistical Process Control

(d) What action, if any, would you recommend

if the next sample was
(i) 196.0, 202.5, 189.0, 197.5, 197.0,
(ii) 204.5, 214.0, 206.0, 207.0, 211.0,
(iii) 188.5, 212.5, 220.0, 215.0, 208.0,
(iv) 206.0, 214.5, 208.5, 209.0, 211.0?
(e) The upper specification limit (USL) is 220 g
and the lower specification limit(LSL) is
200 g Estimate C p where
Cp =

( USL LSL )

.
6
Comment on the ability of the process to
meet this specification.
5. Lengths of cloth produced at a mill often
have to be 'mended' by hand before being
saleable. In sets of 50 the numbers needing
mending were as follows;
17, 14, 13, 16, 14, 16, 22, 19, 15, 16, 12

Set up control charts based on this data.

166

6. A manufacturer of fishing line takes samples

of 60 lengths at regular intervals and counts
the number that break when subjected to a
strain of 38 newtons. During a period when
production was satisfactory the results for
ten samples were as follows:
sample 1
number 14
breaking

2
9

3
7

4 5
15 10

6
8

7 8
13 12

9
6

10
7

(a) Draw a control chart for proportion

defective showing approximate 95% and
99.8% control lines, and plotting the ten
points.
(b) What action would you recommend if the
number breaking in the next sample was
(i) 18
(ii) 24
(iii) 1?
(c) List advantages and/or disadvantages of
the manufacturer's method of control
compared to measuring the actual
breaking strength and setting up control
charts for mean and range.
(AEB)