2/22/2012

Chapter 4

Control Charts for

Measurements with Subgrouping
(for One Variable)

Presumptions

• Subgroups (samples) of data are formed.

• Measurements are made and the values are
obtained with sufficient speed.

1
 2/22/2012

4.1 Basic Control Chart Principles

• Stage 1: Control charts can be used to determine if

a process has been in a state of statistical control
by examining past data. (retrospective data
analysis)
• Stage 2: Recent data can be used to determine
control limits that would apply to future data
obtained from a process.

Basic Control Chart Principles

• Control charts alone cannot produce statistical

control
• Control charts can indicate whether or not
statistical control is being maintained and provide
users with other signals from the data
– To avoid unnecessary and undesired process
adjustments
• Control charts can also be used to study process
capability (Chapter 7)

2
 2/22/2012

Basic Control Chart Principles

• Best results will generally be obtained when

control charts are applied primarily to process
variables than to product variables.
• In general, it is desirable to monitor all process
variables that affect important product variables.
• Control charts are essentially plots of data over
time.

Figure 4.1

3
 2/22/2012

Basic Control Chart Principles

Procedures for setting up new control charts

1. Obtain at least 20 subgroups or at least 100
individual observations (from past or current data)
2. Calculate the trial control limits
3. Data points outside the trial limits should be
investigated. Those points can only be removed if
the assignable causes can be detected and removed
4. Repeat steps 2 and 3 until no further action can be
taken

Basic Control Chart Principles

• After a process has been brought into a state of

statistical control, a process capability study can be
initiated to determine the capability of the process in
regard to meeting the specifications.
– Process performance indices: using long term sigma (when a
process is not in a state of statistical control)

4
 2/22/2012

4.2 Real-time Control Charting vs

Analysis of Past Data
• When a set of points is plotted all at once, the
probability of observing at least one point that is
outside the control limit will be much greater than
.0027.
– When points are plotted individually,
– 3-sigma limits are used,
– A normal distribution is assumed, and
– Parameter of the appropriate distribution are assumed to be known

Table 4.1 Probabilities of Points

Plotting Outside Control Limits
n .0027n Actual Prob.
1 0.0027 0.0027
2 0.0054 0.0054
5 0.0135 0.0134
10 0.0270 0.0267
15 0.0405 0.0397
20 0.0540 0.0526
25 0.0675 0.0654
50 0.1350 0.1264
100 0.2700 0.2369
350 0.9450 0.6118

5
 2/22/2012

Real-time Control Charting vs

Analysis of Past Data
• The calculation of probabilities can only be done if the
parameters are assumed to be known.
• In general, the possible use of k-sigma limits in Stage
1 needs to be addressed.
– Costs of shutting down the process
– Costs of making units outside specifications
– Costs of false looking for assignable causes
• Medical practitioners often tend to favor 2-sigma limits.
(assignable causes can be detected as quickly as
possible)

4.3 Control Charts: When to Use,

Where to Use, How many to Use
Where:
• Not at every work station
• Nature of the product often preclude measurements
• No need for control charts in a process that is highly
unlikely the process ever go out of control
• Control charts should be used where trouble is likely to
occur
• They should be used where the potential for cost reduction
is substantial
Number of charts
• Computer vs manual charting

6
 2/22/2012

4.4 Benefits from the Use of

Control Charts
• Good record keeping
• Control charts work as an aid in identifying special causes
of variation

4.5 Rational Subgroups

• Important Requirement: data chosen for the subgroups

come from the same population (same operator, shift,
machine, etc.)
• If data are mixed, the control limits will correspond to a
mixture (aggregated) distribution

7
 2/22/2012

4.6 Basic Statistical Aspects of

Control Charts
• If we are interested in controlling the process mean, ,
using the subgroup mean (𝑋), and the limits are given as
𝜇 ± 3𝜎𝑋 , the total probability outside the limits is .0027 if X
has a normal distribution.
– Exactly a normal distribution
– 𝜎𝑋 is known
– Process mean is at 
• 3-sigma limits (rather than probability limits)
• The distribution of 𝑋 will be more normal than the
distribution of X (Central Limit Theorem)

Basic Statistical Aspects of

Control Charts
• When X is highly asymmetric, data can usually be
transformed (log, square root, power, reciprocal, Cox-cox,
etc.) to be approximately normal.

8
 2/22/2012

4.7 Illustrative Example

• Important Requirement: data chosen for the subgroups

come from the same population (same operator, shift,
machine, etc.)
• If data are mixed, the control limits will correspond to a
mixture (aggregated) distribution

Table 4.2 Data in Subgroups

Obtained at Regular Intervals
Subgroup X1 X2 X3 X4 X-bar R S
1 72 84 79 49 71.00 35 15.47
2 56 87 33 42 54.50 54 23.64
3 55 73 22 60 52.50 51 21.70
4 44 80 54 74 63.00 36 16.85
5 97 26 48 58 57.25 71 29.68
6 83 89 91 62 81.25 29 13.28
7 47 66 53 58 56.00 19 8.04
8 88 50 84 69 72.75 38 17.23
9 57 47 41 46 47.75 16 6.70
10 13 10 30 32 21.25 22 11.35
11 26 39 52 48 41.25 26 11.53
12 46 27 63 34 42.50 36 15.76
13 49 62 78 87 69.00 38 16.87
14 71 63 82 55 67.75 27 11.53
15 71 58 69 70 67.00 13 6.06
16 67 69 70 94 75.00 27 12.73
17 55 63 72 49 59.75 23 9.98
18 49 51 55 76 57.75 27 12.42
19 72 80 61 59 68.00 21 9.83
20 61 74 62 57 63.50 17 7.33

9
 2/22/2012

R vs. S Charts

• Either R or S charts could be used in controlling the

process variability.
• S-chart is preferable since it uses all the observations in
each subgroup.
• Other statistical methods in quality improvement are
generally based on S (or S2)

Estimating of Population Parameters

by Sample Statistics
• Population Statistics:  , usually unknown
• Using Sample Statistics to estimate population
statistics:
– Point estimates ̂  x  x R
ˆ  ˆ 
s
k d2 c4

10
 2/22/2012

4.7.1 R-Chart

• 3-sigma limits
𝑅 ± 3𝜎𝑅 (4.1)

• 𝐿𝐶𝐿 = 𝐷3 𝑅
• 𝑈𝐶𝐿 = 𝐷4 𝑅

• Assumption of normality
• LCL for R-chart will be zero when the subgroup size is less
than 7

Figure 4.2 R-Chart

11
 2/22/2012

4.7.2 R-Chart with Probability Limits

• Probability limits (Table F)

𝑅
• 𝑈𝐶𝐿 = 𝐷.999 (𝑑 )
2

𝑅
• 𝐿𝐶𝐿 = 𝐷.001 (𝑑 )
2

• LCL for R-chart always exist

• Asymmetry
• Assumption of normality of X

4.7.3 S-Chart

• 3-sigma limits
𝑆 ± 3𝜎𝑆 (4.2)

• 𝐿𝐶𝐿 = 𝐵3 𝑆
• 𝑈𝐶𝐿 = 𝐵4 𝑆

• Assumption of normality
• LCL for R-chart will be zero when the subgroup size is less
than 6

12
 2/22/2012

S-Chart

4.7.4 S-Chart with Probability Limits

(𝑛−1)𝑆 2
• If X ~ N(, 2), then ~ 𝜒 2 𝑛−1
𝜎2
𝑛 − 1 𝑆2
𝑃 𝜒 2 .001 < < 𝜒 2 .999 = .998
𝜎2

𝜎2 2 𝜎2
𝑃 𝜒 < 𝑆2 < 𝜒2 = .998
𝑛 − 1 .001 𝑛 − 1 .999

𝜒 2 .001 𝜒 2 .999
𝑃 𝜎 <𝑆<𝜎 = .998
𝑛−1 𝑛−1

𝜒2 .001 𝜒2 .999 𝜒2 .001 𝜒2 .999

LCL=𝜎 𝑛−1
UCL=𝜎 𝑛−1
or LCL=𝑐𝑆 𝑛−1
UCL= =𝑐𝑆 𝑛−1
4 4

13
 2/22/2012

Table 4.3 The .001 and .999

Percentage Points of 2 Distribution
n 𝜒 2 .001 𝜒 2 .999
2 1.5708E-06 10.8276
3 0.0020 13.8155
4 0.0243 16.2662
5 0.0908 18.4668
6 0.2102 20.5150
7 0.3811 22.4577
8 0.5985 24.3219
9 0.8571 26.1245
10 1.1519 27.8772
11 1.4787 29.5883
12 1.8339 31.2641
13 2.2142 32.9095
14 2.6172 34.5282
15 3.0407 36.1233

S-Chart with Probability Limits

• LCL for S-chart always exist

• Asymmetry
• Assumption of normality of X
• Assumption of  = 𝑆/𝑐4
• With 3-sigma limits, the number of “false alarms” (1/.0045
=222) will be greater than expected (1/.00135=741)

14
 2/22/2012

4.7.5 S2-Chart

2
𝜒 2 .001
𝐿𝐶𝐿 = 𝑆
𝑛−1

2
𝜒 2 .999
𝑈𝐶𝐿 = 𝑆
𝑛−1
Where 𝑆 2 is the average of the S2 values

4.7.6 𝑋-Chart

• All of the points for data in Table 4.2 lie within the control
limits (R, S-Charts). The process variability is evidently in a
state of statistical control.

𝑅
𝜎𝑋 𝑑2
𝑋 ± 3𝜎𝑋 = 𝑋 ± 3 =𝑋±3 = 𝑋 ± 3𝐴2 𝑅 (4.3)
𝑛 𝑛
𝜎𝑋 𝑆
𝑐4
𝑋 ± 3𝜎𝑋 = 𝑋 ± 3 =𝑋±3 = 𝑋 ± 3𝐴3 𝑆
𝑛 𝑛

15
 2/22/2012

Figure 4.3 𝑋-Chart

4.7.6 𝑋-Chart

Questions to ask:
• What is the likelihood to have one of 20 sample data fall
outside the control limits?
• What is the probability for an subgroup average as small as
21.25?

16
 2/22/2012

4.7.6 𝑋-Chart

Estimation of :
• 𝑆/𝑐4
• 𝐴𝑣𝑒(𝑆 2 )
• 𝑐4 (𝑚𝑛 − 𝑚 + 1) 𝐴𝑣𝑒(𝑆 2 )

4.7.7 Recomputing Control Limits

• If the cause was detected as operator error, but there is no way

to determine the correct value(s), the control limits should be
recomputed using the remaining 19 subgroups.
• When the original data set contains m subgroups and one of the
subgroup averages is below the LCL, it can be shown that the
new UCL must exceed the old UCL by at least
1
𝐴3 𝑆 𝑛 , provided that the S of the deleted subgroup is
𝑛−1
less than 𝑆 𝑛 .
• If 30 samples of size 4 are used to compute the control limits for
an R-chart, and one of the ranges is above the UCL, the smallest
possible difference between the original and recomputed UCL
value is .101𝑅.

17
 2/22/2012

4.7.8 Applying Control Limits to

Future Production
• If the cause was detected as operator error, but there is no way
to determine the correct value(s), the control limits should be
recomputed using the remaining 19 subgroups.
• When the original data set contains m subgroups and one of the
subgroup averages is below the LCL, it can be shown that the
new UCL must exceed the old UCL by at least
1
𝐴3 𝑆 𝑛 , provided that the S of the deleted subgroup is
𝑛−1
less than 𝑆 𝑛 .
• If 30 samples of size 4 are used to compute the control limits for
an R-chart, and one of the ranges is above the UCL, the smallest
possible difference between the original and recomputed UCL
value is .101𝑅.

18