ME 472 Engineering Metrology

and Quality Control

Dr. A. Tolga Bozdana
Assistant Professor
Mechanical Engineering
University of Gaziantep
Statistical Quality Control and
Use of Control Charts
1
Statistical Process Control (SPC) & Control Charts
The most frequently used charts are Mean and Range Charts, which are used together.
SPC is achieved with the aid of control charts (pioneered by Walter A. Shewhart in 1920).
Achievement of SPC
The process capability for specified tolerances is examined using capability indices.
If the process is not in control, efforts are made for making the process under control.
Control limits are determined in order to check stability of the process.
The variable (length, volume, weight, time, etc.) is measured for each item of the sample so that
the sample mean and range values are recorded on a chart.

Periodically, samples of a given size (four steel rods, five tins of paint, eight tablets, four delivery
times, etc.) are taken from the process at reasonable intervals, when it is believed to be stable or
in-control and adjustments are not being made.

The operation of control charts to detect the state of control of a process is as follows:
Implementation of Control Charts
To control a process based on varying data, it is necessary to keep a check on the current state of
accuracy (central tendency) and precision (spread) of the distribution of data.

SPC uses statistical tools to observe the performance of production process in order to predict
significant deviations that may later result in rejected product.

Definition of SPC
2
A Case Study
151 157 140 154 25
149 157 152 151 24
151 151 156 143 23
154 158 147 152 22
160 148 156 146 21
150 150 142 155 20
151 149 155 153 19
153 149 148 157 18
148 153 150 155 17
149 148 144 147 16
157 148 146 150 15
149 150 160 144 14
142 152 146 152 13
148 152 145 155 12
153 154 150 151 11
154 152 148 145 10
156 149 150 158 9
155 149 147 141 8
155 137 144 149 7
147 145 150 157 6
157 155 153 157 5
148 152 146 154 4
152 143 139 145 3
153 134 150 151 2
146 154 146 144 1
iv iii ii i
Rod Lengths (mm)
Sample
No.
Lets represent this data using a histogram chart, which is used
for showing the frequency of data values. First, we must define
the bins (i.e. intervals).
For this purpose, we search for the min. and the max. values
in the data set (which are 134 and 160).
151 157 140 154 25
149 157 152 151 24
151 151 156 143 23
154 158 147 152 22
160 148 156 146 21
150 150 142 155 20
151 149 155 153 19
153 149 148 157 18
148 153 150 155 17
149 148 144 147 16
157 148 146 150 15
149 150 160 144 14
142 152 146 152 13
148 152 145 155 12
153 154 150 151 11
154 152 148 145 10
156 149 150 158 9
155 149 147 141 8
155 137 144 149 7
147 145 150 157 6
157 155 153 157 5
148 152 146 154 4
152 143 139 145 3
153 134 150 151 2
146 154 146 144 1
iv iii ii i
Rod Lengths (mm)
Sample
No.
Then, we define starting & end values for the bins (i.e. 133.5 &
160.5). After this, we divide the difference between these values
by equal intervals (i.e. 27/3 which gives 9 equal intervals).
Therefore, the bins for this data set can be:
145.5 148.5
142.5 145.5
139.5 142.5
136.5 139.5
133.5 136.5
157.5 160.5
154.5 157.5
151.5 154.5
148.5 151.5
Suppose that, for production of rods, we have 100 rod lengths
as 25 samples of size 4 were taken.
3
0
2
4
6
8
10
12
14
16
18
20
22
24
26
1
3
5
1
3
8
1
4
1
1
4
4
1
4
7
1
5
0
1
5
3
1
5
6
1
5
9
Midpoint
F
r
e
q
u
e
n
c
y
Histogram Chart
Select the data set (i.e. 100 rod
lengths) and the intervals so that
the histogram chart will be built.
The frequency of each value within
the data set according to each bin
will be calculated automatically.
Sometimes, it is more practical to
show midpoints in the chart rather
than intervals. In such cases, the
midpoint of each bin is calculated
and used in the chart (e.g. mean
of interval 133.5 136.5 is 135).
Histogram charts are also useful for
showing the distribution of data sets.
0
2
4
6
8
10
12
14
16
18
20
22
24
26
1
3
5
1
3
8
1
4
1
1
4
4
1
4
7
1
5
0
1
5
3
1
5
6
1
5
9
Midpoint
F
r
e
q
u
e
n
c
y
1
3
3
.
5
1
3
6
.
5
1
3
9
.
5
1
4
2
.
5
1
4
5
.
5
1
4
8
.
5
1
5
1
.
5
1
5
4
.
5
1
5
7
.
5
1
6
0
.
5
Intervals
The histogram chart is obtained using Microsoft Excel software (Click on the Data Analysis under
Tools menu and choose Histogram option from the list. If you cant see Data Analysis function,
then go to Add-ins under Tools menu and click on Analysis Toolpak).
4
Sample Mean & Grand (Process) Mean
Instead, Grand (Process) Mean (i.e. the mean of all
sample means) is a good estimate of population mean:
mm 1 . 150 =

=
=
k
i
i
k X X
1
First, the mean of each sample is calculated, e.g. mean
of sample no. 9 is:
( ) mm X 25 . 153 4 156 149 150 158 = + + + =
Then, the process mean is calculated as:
( )
mm X 1 . 150
25
50 . 150 ... 75 . 144 00 . 147 50 . 147
=
+ + + +
=
151 157 140 154 25
149 157 152 151 24
151 151 156 143 23
154 158 147 152 22
160 148 156 146 21
150 150 142 155 20
151 149 155 153 19
153 149 148 157 18
148 153 150 155 17
149 148 144 147 16
157 148 146 150 15
149 150 160 144 14
142 152 146 152 13
148 152 145 155 12
153 154 150 151 11
154 152 148 145 10
156 149 150 158 9
155 149 147 141 8
155 137 144 149 7
147 145 150 157 6
157 155 153 157 5
148 152 146 154 4
152 143 139 145 3
153 134 150 151 2
146 154 146 144 1
iv iii ii i
Rod Lengths (mm)
Sample
No.
151 157 140 154 25
149 157 152 151 24
151 151 156 143 23
154 158 147 152 22
160 148 156 146 21
150 150 142 155 20
151 149 155 153 19
153 149 148 157 18
148 153 150 155 17
149 148 144 147 16
157 148 146 150 15
149 150 160 144 14
142 152 146 152 13
148 152 145 155 12
153 154 150 151 11
154 152 148 145 10
156 149 150 158 9
155 149 147 141 8
155 137 144 149 7
147 145 150 157 6
157 155 153 157 5
148 152 146 154 4
152 143 139 145 3
153 134 150 151 2
146 154 146 144 1
iv iii ii i
Rod Lengths (mm)
Sample
No.
150.50 151 157 140 154 25
152.25 149 157 152 151 24
150.25 151 151 156 143 23
152.75 154 158 147 152 22
152.50 160 148 156 146 21
149.25 150 150 142 155 20
152.00 151 149 155 153 19
151.75 153 149 148 157 18
151.50 148 153 150 155 17
147.00 149 148 144 147 16
150.25 157 148 146 150 15
150.75 149 150 160 144 14
148.00 142 152 146 152 13
150.00 148 152 145 155 12
152.00 153 154 150 151 11
149.75 154 152 148 145 10
153.25 156 149 150 158 9
148.00 155 149 147 141 8
146.25 155 137 144 149 7
149.75 147 145 150 157 6
155.50 157 155 153 157 5
150.00 148 152 146 154 4
144.75 152 143 139 145 3
147.00 153 134 150 151 2
147.50 146 154 146 144 1
iv iii ii i
Sample
Mean
(mm)
Rod Lengths (mm)
Sample
No.
For the given data, it is not meaningful to calculate the
mean of a whole population (i.e. the total output from a
process rather than a sample):
5
Standard Error (Deviation) of Means
Population of
individuals, x
Distribution of
sample means,
(sample size n)
X
F
r
e
q
u
e
n
c
y
F
r
e
q
u
e
n
c
y

X
SE

Based on the information in previous slides, it is

now better to use the standard deviation of
sample means instead of using standard
deviation of the population.
Standard Error (SE) of Means is defined as:
n SE =
where is the standard deviation of population,
n is the sample size (total number of samples).
In case of our example, n is 4.
SE of means is more reliable than SD of whole
population. As seen from the graph, the scatter
(i.e. the precision) of the sample means is much
less than the scatter of the individual rod lengths.
On the other hand, the mean (i.e. the accuracy)
remains unchanged as population is the same:
mm X 1 . 150 = =
6
Range (Measure of Precision or Spread)
150.50 151 157 140 154 25
152.25 149 157 152 151 24
150.25 151 151 156 143 23
152.75 154 158 147 152 22
152.50 160 148 156 146 21
149.25 150 150 142 155 20
152.00 151 149 155 153 19
151.75 153 149 148 157 18
151.50 148 153 150 155 17
147.00 149 148 144 147 16
150.25 157 148 146 150 15
150.75 149 150 160 144 14
148.00 142 152 146 152 13
150.00 148 152 145 155 12
152.00 153 154 150 151 11
149.75 154 152 148 145 10
153.25 156 149 150 158 9
148.00 155 149 147 141 8
146.25 155 137 144 149 7
149.75 147 145 150 157 6
155.50 157 155 153 157 5
150.00 148 152 146 154 4
144.75 152 143 139 145 3
147.00 153 134 150 151 2
147.50 146 154 146 144 1
iv iii ii i
Sample
Mean
(mm)
Rod Lengths (mm)
Sample
No.
The range (i.e. difference between the highest
and the lowest observations) is the simplest
possible measure of scatter:

=
=
k
i
i
k R R
1
For instance, the range in sample no. 9 is the
difference between the longest (158 mm) and
the shortest (149 mm), which is 9 mm. Hence,
the mean range (i.e. the mean of all sample
ranges) is found as:
17 150.50 151 157 140 154 25
8 152.25 149 157 152 151 24
13 150.25 151 151 156 143 23
11 152.75 154 158 147 152 22
14 152.50 160 148 156 146 21
13 149.25 150 150 142 155 20
6 152.00 151 149 155 153 19
9 151.75 153 149 148 157 18
7 151.50 148 153 150 155 17
5 147.00 149 148 144 147 16
11 150.25 157 148 146 150 15
16 150.75 149 150 160 144 14
10 148.00 142 152 146 152 13
10 150.00 148 152 145 155 12
4 152.00 153 154 150 151 11
9 149.75 154 152 148 145 10
9 153.25 156 149 150 158 9
14 148.00 155 149 147 141 8
18 146.25 155 137 144 149 7
12 149.75 147 145 150 157 6
4 155.50 157 155 153 157 5
8 150.00 148 152 146 154 4
13 144.75 152 143 139 145 3
19 147.00 153 134 150 151 2
10 147.50 146 154 146 144 1
iv iii ii i
Sample
Range
(mm)
Sample
Mean
(mm)
Rod Lengths (mm)
Sample
No.
( )
mm R 8 . 10
25
17 ... 13 19 10
=
+ + + +
=
It is possible to simplify the calculation of warning
and action limits. In statistical process control for
variables, the sample size is usually less than 10;
so it is possible to use an alternative measure of
spread of process: the mean range of samples.
7
Building X-Bar (Mean) Chart
In a stable process, most of the sample means lie within the range of . So, we extrapolate
the process mean by on both sides in order to obtain upper and lower action limits. If the
process is running satisfactorily, we expect from the normal distribution that at least 0.27% ( 1 in 370)
of the means of successive samples should lie out of action limits.
SE X 3
SE 3
We can extrapolate process mean by
to obtain upper and lower
warning limits. Thus, at least 4.45%
( 1 in 22) of the means of successive
samples should lie out of warning limits.
SE 2
stable zone
warning zone
action zone
stable zone
warning zone
action zone
Chance of having two consecutive
sample means in the warning zone is
about 1/22 x 1/22 = 1/484, which is
even lower than the chance of one
value in action zone, suggests that
the process is out of control.
First of all, we draw the distribution of sample means graph, and then turn this bell-shape graph onto
its side having process mean in x-axis and sample mean in y-axis.
8
Calculation of Action and Warning Limits
2
/ / d R d R
n
= =
Hence, when the limits are being calculated, the mean range can be used instead of SE:
R
n d
X
n
X SE X
n
3
3 3 Limits Action = = =

R
n d
X
n
X SE X
n
2
2 2 Limits Warning = = =

As d
n
and n are all constants for the same sample size, it is possible to replace numbers and symbols
with just one constant, called A
2
:
2 2
3 2
2
&
3
A
n d
A
n d
n n
= =
As a result, action and warning limits can be written in terms of the mean range and a constant:
R A X
2
Limits Action = R A X
2
3 2 Limits Warning = &
Standard deviation can be expressed in terms of the mean range and Hartleys constant (d
n
or d
2
) :
9
Constants for Mean Charts
0.27
0.29
0.31
0.34
0.37
0.42
0.48
0.58
0.73
1.02
1.88
A
2
Constant for use with
Sample Range
3.258
3.173
3.078
2.970
2.847
2.704
2.534
2.326
2.059
1.693
1.128
Hartleys Constant
(d
n
or d
2
)
0.18 12
0.19 11
0.21 10
0.20 9
0.25 8
0.28 7
0.32 6
0.39 5
0.49 4
0.68 3
1.25 2
2/3 A
2
Sample
Size (n)
The constants (d
n
and A
2
) for sample
sizes (n) from 2 to 12 are available.
For the sample sizes up to 12, the range
method of estimating action and warning
limit is relatively efficient.
For the values greater than 12, the range
loses its efficiency rapidly as it ignores all
the information in the sample between the
highest and the lowest values.
The sample sizes of 4 or 5 are generally
employed in control charts, which gives
satisfactory results.
In case of our example, the sample size
(n) is 4. Therefore, the constants (d
n
and
A
2
) are selected accordingly.
0.27
0.29
0.31
0.34
0.37
0.42
0.48
0.58
0.73
1.02
1.88
A
2
Constant for use with
Sample Range
3.258
3.173
3.078
2.970
2.847
2.704
2.534
2.326
2.059
1.693
1.128
Hartleys Constant
(d
n
or d
2
)
0.18 12
0.19 11
0.21 10
0.20 9
0.25 8
0.28 7
0.32 6
0.39 5
0.49 4
0.68 3
1.25 2
2/3 A
2
Sample
Size (n)
10
Case Study Mean Chart
Process Mean
Then, action and warning limits are determined based on mean range and constants:
059 . 2 d : table) (from Constant s Hartley'
8 . 10 : d) (calculate Range Mean
n
=
= mm R
49 . 0 A 3 2 & 73 . 0 A
: table) (from Constants Range Sample
2 2
= =
mm R A X
mm R A X
22 . 142 : Limit Action Lower
98 . 157 : Limit Action Upper
2
2
=
= +
mm R A X
mm R A X
81 . 144 3 2 : Limit ing Lower Warn
40 . 155 3 2 : Limit ing Upper Warn
2
2
=
= +
140
142
144
146
148
150
152
154
156
158
160
S
a
m
p
l
e

M
e
a
n
140
142
144
146
148
150
152
154
156
158
160
S
a
m
p
l
e

M
e
a
n
Upper Warning Limit
Upper Action Limit
Lower Action Limit
Lower Warning Limit
The mean chart is constructed by turning the bell-shape histogram chart onto its side, hence we have
sample mean as y-axis and the process mean as x-axis (which was calculated before as 150.1 mm).
11
Building R-bar (Range) Chart
The control limits on a range chart are
asymmetrical about the mean range
since the distribution of sample ranges
is a positively skewed distribution.
0.1 %
F
r
e
q
u
e
n
c
y
Sample Range
0.1 %
2.5 %
2.5 %
R
1.87
1.91
1.93
1.99
2.04
2.11
2.21
2.34
2.57
2.98
4.12
D
0.001
0.40
0.38
0.35
0.32
0.29
0.26
0.21
0.16
0.10
0.04
0.00
D
0.999
0.58
0.56
0.54
0.52
0.50
0.46
0.42
0.37
0.29
0.18
0.04
D
0.975
Constant for use with Mean Range
1.51 12
1.53 11
1.56 10
1.58 9
1.62 8
1.66 7
1.72 6
1.81 5
1.93 4
2.17 3
2.81 2
D
0.025
Sample
Size (n)
Thus, upper and lower action limits are
calculated for 0.1% & 99.9% of sample
range while upper and lower warning
limits lie at 2.5% & 97.5%:
R D
R D
R D
R D
.
.
.
.
975 0
025 0
999 0
001 0
: Limit ing Lower Warn
: Limit ning Upper War
: Limit Action Lower
: Limit Action Upper

The constants for calculating control

limits are available for sample sizes
(n) up to 12.
For our case, the constants are
selected from the table for sample
size of 4.
1.87
1.91
1.93
1.99
2.04
2.11
2.21
2.34
2.57
2.98
4.12
D
0.001
0.40
0.38
0.35
0.32
0.29
0.26
0.21
0.16
0.10
0.04
0.00
D
0.999
0.58
0.56
0.54
0.52
0.50
0.46
0.42
0.37
0.29
0.18
0.04
D
0.975
Constant for use with Mean Range
1.51 12
1.53 11
1.56 10
1.58 9
1.62 8
1.66 7
1.72 6
1.81 5
1.93 4
2.17 3
2.81 2
D
0.025
Sample
Size (n)
12
0
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30
S
a
m
p
l
e

R
a
n
g
e
Case Study Range Chart
0
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30
S
a
m
p
l
e

R
a
n
g
e
Mean Range
Upper Warning Limit
Upper Action Limit
Lower Action Limit
Lower Warning Limit
Then, action and warning limits are determined based on mean range and constants:
29 . 0
93 . 1
10 . 0
57 . 2
975 0
025 0
999 0
001 0
=

.
.
.
.
D
D
D
D
f
r
o
m
t
a
b
l
e
mm R D
mm R D
.
.
1 . 1 : Limit Action Lower
8 . 27 : Limit Action Upper
999 0
001 0
=

mm R D
mm R D
.
.
1 . 3 : Limit ing Lower Warn
8 . 20 : Limit ning Upper War
975 0
025 0
=

The range chart is constructed by turning the bell-shape histogram chart onto its side, hence we have
sample range as y-axis and mean range as x-axis (which was calculated before as 10.8 mm).
13
Is the process in control?
140
142
144
146
148
150
152
154
156
158
160
S
a
m
p
l
e

M
e
a
n
0
3
6
9
12
15
18
21
24
27
30
S
a
m
p
l
e

R
a
n
g
e
NO mean or range values which lie outside the action limits (action zone)?

NO more than about 1 in 22 values between warning and action limits (warning zone)?

NO incidence of two consecutive mean or range values which lie outside the same warning limit on either
mean or range chart (warning zone)?

NO run or trend of five or more consecutive mean or range values, which also infringes a warning or action
limit (warning zone and/or action zone)?

NO run of more than six sample means which lie either above or below the process mean (stable zone)?

NO trend of more than six values of the sample means which are either rising or falling (stable zone)?

MEAN CHART RANGE CHART

(should be examined first)
PROCESS IS IN CONTROL !!!
14
What if the process is out of control?
Do not use the control charts and investigate the assignable causes of variation.
The assignable causes of variation have been identified, and now to be eliminated.
Another set of samples from
the process is taken and control
limits are recalculated.
Approximate control limits are recalculated by simply
excluding the out of control results for which special
causes have been found and corrected.
The exclusion of samples representing unstable
conditions is not just throwing away bad data.
By excluding the data affected by known causes,
we have a better estimate of variation due to
common causes only.
The process is re-examined to see if it is in statistical control.
If the process is shown to be in statistical control, the next task is to
compare the limits of this control with the required tolerance.
15
Process Variability and Tolerances
The relationship between process variability and tolerances can be formalized by consideration of
standard deviation of the process. In order to manufacture within the specification, distance between
Upper Specification Limit (USL) or Upper Tolerance (+T) and Lower Specification Limit (LSL)
or Lower Tolerance (T) must be analyzed.
Low Relative Precision: 2T < 6
Medium Relative Precision: 2T 6
High Relative Precision: 2T >> 6
There are three precision levels:
A comparison of process capability and tolerance can only take place, with confidence, when
the process is statistically in control. Thus, in controlling a process, it is necessary to establish
first that it is in statistical control and then to compare its centring & spread with the specified
target value and the specification tolerance.

A process may not meet the specification requirements, but still be in a state of statistical control.
Tolerance limits should be based on the functional requirements of the product.
Limits on control charts are based on the stability and actual capability of the process.

Control Limits vs. Tolerance Limits

16
Process Capability Indices
possible complications of non-normal distributions
detection of changes in the scatter
possible changes in the scatter (range)
detection of changes of the mean
some movement of the mean
Calculation of process capability indices estimates the short-term variations within the process.
This short-term is the period over which the process remains relatively stable. However, we know
that processes do not remain stable for all time, and therefore we need to allow within the specified
tolerance limits for:
Process capability indices are simply a means of indicating the variability of a process relative
to the product specification tolerance.

It is a measure relating the actual performance of a process to its specified performance,

where the processes are considered to be a combination of plant or equipment, the method itself,
the people, the materials, and the environment.

Process Capability Index

17
Relative Precision Index (RPI)
We know that:
Constant s y Hartle
Ranges Sample of Mean

= =
n
d
R
Then, we obtain:
n
d R
T 6 2

In our case study: 914 . 2

059 . 2
6 6
RPI Min. = = =
n
d
If we are asked to produce
rods within 10 mm of the
target length:
2T = 20 mm 852 . 1
8 . 10
20 2
RPI = = =
R
T
Reject
Material
If the specified tolerances
are widened to 20 mm:
2T = 40 mm 704 . 3
8 . 10
40 2
RPI = = =
R
T
Avoid
Rejecting
Material
is known as Relative Precision Index (RPI) and the value of is the minimum RPI
to avoid the production of a material outside the specification limits. Moreover, RPI index does not
comment on the centring of a process as it deals only with its relative spread or variation.
R T / 2
n
d / 6
This is the oldest index based on a ratio of the mean range of samples to the tolerance band.
In order to avoid the production of defective material, the specification width must be greater than
the process variation:
6 2 T
18
C
p
Index
6
2
6
T LSL USL
C
p
=

=
Process variation is greater than tolerance band, so the process is incapable.
(a) C
p
< 1
For increasing values of C
p
, the process becomes increasingly capable. (b) C
p
1
Similar with RPI, C
p
index makes no comment about the centring of the process, which means
that it is a simple comparison of total variation with tolerances.
In order to manufacture within a specification, difference between USL and LSL must be less than
total process variation. So, a comparison of 6 with (USL - LSL) or 2T gives C
p
index:
19
C
pk
Index
A situation like this invalidate the use of C
p
index, hence
there is a need for another index which takes account of
both the process variation and the centring. Such an
index is called C
pk
index, which is widely accepted as a
means of communicating process capability.

= =
3
C or
3
C of lesser
l u
pk pk
LSL X X USL
C
pk
For USL and LSL limits, there are two C
pk
values (C
pk
u
and C
pk
l
). These relate the difference between
process mean and USL/LSL respectively, and the lesser of these two will be the value of C
pk
index:
C
pk
can be used when there is only one specification limit, but C
p
cannot be used in such cases.
Comparison of C
p
and C
pk
shows no difference if the process is centred on the target.
It may be possible to increase C
pk
value by centring the process so that its mean value coincides
with the mid-specification (target).

As in the case of C
p
, increasing values of C
pk
correspond to increasing capability.
C
pk
< 1 means that, considering the process variation and its centring, at least one of the tolerance
limits is exceeded and the process is incapable.