Chapter - 6

Statistical Process Control using

Control Charts

1
Introduction to Control Charts
 Emphasis is on process control and improvement
 A Control chart is a
 Graphical tool
 For monitoring the activity of an ongoing process
 It is called as ‘Shewhart’ chart
 Values of quality characteristic is plotted along
‘ordinate’
 Sample number or subgroups (in order of time) is
plotted along ‘abscissa’
 Example for quality characteristic includes Average
length, Average tensile strength, Average service
time etc.
 Quality characteristic can be categorized as
 Variable: Numerical values can be obtained

 Attribute: No numerical values, only conforming or2

non conforming
Control Chart – Contd..

3
 More about Control Charts –
Contd..
 A Control chart has
 A centre line, represents the average value of the
characteristic
 It indicates, where the process is centred
 UCL and LCL, used to make decision regarding the process
 If the points plot within the control limits and do not exhibit
any identifiable pattern, then process is in statistical control
 If a point plots outside the control limits or if an identifiable
random pattern exists process is out of statistical control
 Benefits
 Indicates when something may be wrong, so that corrective
action can be taken
 Pattern of plot can help in diagnosing the possible causes
and hence possible remedial action
 Helps in estimating the process capability of the process

4
 Causes of Variation
• Variability is part of any process
• Causes can be men, methods, machine, materials, environment
• Can be categorized in to two
 Special causes  Common causes
 Assignable causes or  Chance causes or
sporadic problem chronic problem
 Not inherent in the  Inherent in the process
process  Affects all items, as it is
part of process design
 Will not affect all items
 Exists as long as the
 Can be due to use of a process is not changed
wrong tool, improper  Can be due to vibration
raw material, operator of a machine,
error etc. fluctuation etc.

5
 Causes of Variation – Contd..

 Special causes  Common causes

 If an observation falls  Variation is the effect of
outside the control limits or many small causes and
a non-random pattern
exists, special causes
cannot be totally
exists eliminated
 Control chart is used to  When this variation is
detect the presence of constant, called as stable
special causes and should system of common
be eliminated causes
 15% of all problems are  Management is alone
due to special causes responsible for common
 Action is required from both causes
management and workers
 85% of all problems are
due to common causes
6
Statistical Basis for Control Charts
 Centre line is based on the mean of a
process and is found by taking the
average of the values in sample
 Can also be desirable target or standard
value
 Role of Normal distribution
 Value of the statistic plotted on a control
chart are assumed to have an approximately
normal distribution
 Any sample from population distribution that
is uni-modal and symmetric, the central limit
theorem states that if the plotted statistic is a
sample average, it tend to have a normal
distribution
7
Control Chart – Contd..

8
Statistical Basis – Contd..
 Why 3σ limits?
 Chosen in such a way that probability of the sample points
falling between them is ‘1’, if the process is in statistical
control
 Normal distribution theory states that a sample statistic will
fall within the limits 99.74% of the time, if the process is in
control
 Most common basis for deciding whether a process is
out of control is “the presence of a sample statistic
outside the control limits” and also it depends on other
rules
 A control chart is a means of online process control
 If the control limits are calculated from current data, then
it tells whether the process is presently in control or not
 If the control limits are calculated from previous data,
based on a process that was in control, it helps to find
whether the process has drifted out of control

 Control chart can also estimate process parameters 9

such as mean, standard deviation and fall outs
Statistical Basis for Control Charts
– Contd..
 It can be used to determine the capability of
process i.e, ability of the process to produce
within desirable specifications
 Process capability studies help in making
major management decisions like make or
buy, investment in machines, vendor selection
etc.
 Issues in construction of control chart are:
 Selection of control limits, No. of items in the
sample, Frequency of sampling, Minimizing
error in inferences, Interpretation of control
charts etc.
10
Selection of Control Limits
 Let ‘θ’ is the mean diameter of parts produced by
the process and θ* would be sample mean
diameter of parts chosen from the process
 E(θ*) is the mean or expected value and σ(θ*) be the
standard deviation of the estimator θ*
 Then the centre line and control limits are given
by
 CL = E(θ*)
 UCL = E(θ*) + k σ(θ*)
 LCL = E(θ*) - k σ(θ*)
 Where k = no. of standard deviations of the
sample statistic that the control limits are
placed from the centre line
 Generally value of k is chosen as 3
11
Selection of Control Limits –
Contd..
 Sample statistic is assumed to have an normal
distribution
 Value of k = 3 implies that there is a probability of
only 0.0026 of a sample statistic falling outside
the control limits, if the process in in control
 k is also chosen, based on the desired probability
of sample statistic falling outside the control
limits, when process is in control. These limits
are called “Probability limits”
 Choice of k is influenced by the Error
consideration (Type I error and Type II error)

12
Errors in making inferences
 Making inferences from a control chart is similar to testing a
hypothesis
 Control limits are the critical points that separate the
rejection and acceptance region
 If a sample values fall above UCL or below LCL, then we
reject null hypothesis or say that process is out of control
 Type I error
 Inferring that a process is out of control, when it is actually
in control
 Probability of Type I error is given by ‘α’
 Type II error
 Inferring that a process is in control, when it is really out of
control
 All points may fall within limit, but process mean has
changed

13
Operating Characteristics curve
(OC Curve)
 An OC curve is a measure of goodness of a control chart’s
ability to detect changes in process parameters
 A plot between the probability of the Type II error versus
the shifting of a process parameter value from its in-control
value
 Helps us to determine the chances of not detecting a shift
of a certain magnitude in a process parameter on a control
chart
 Shape of the curve is similar to inverted ‘S’
 For small shifts in the process mean, the probability of non-
detection is high
 If the change in process mean is high, the probability of
detection increases
 The ability of the control chart to detect changes quickly is
indicated by the steepness of the OC-curve and the
quickness with which the probability of non-detection
reaches zero
14
OC Curve

15
Effect of control limits on errors in
inference making
 Choice of the control limits influence the occurrence
of Type I error and Type II error
 As control limit are placed farther apart, the
probability of Type I error decreases
 For 3σ limit, the probability of Type I error is 0.0026
while for 2σ limit, it is 0.0124
 But probability of making Type II error increases,
showing that both these errors are inversely related
 If all other process parameters held fixed, probability
of Type II error decreases with an increase in sample
size
 As n increases, standard deviation decreases and
hence control limit will be nearer, and reduces
probability of Type II error.

16
 Effect of control limits on errors in
inference making – Contd..
 It involves making a judicious choice of control limit
selection
 Generally, control limits are placed at 3σ distance from
centre line, restricting the occurrence of ‘false alarm’
 For critical process, it is often necessary to detect
small changes as early as possible, hence tighter
control limits are preferable
 Warning limits
 Are usually placed at 2σ limits from the centre line
 When a point falls outside the warning limits, but within
control limits, process ‘may’ go out of control
 For a normally distributed statistic, the probability of falling
in a band between the warning and control limit is 4.3%

17
Effect of sample size on control
limits
 Sample size affects the standard deviation of the sample
statistic plotted
 In a control chart, for the sample mean standard deviation
is given by σx = σ / √n
 Standard deviation and sample size are inversely related
 n increases, σ decreases, i.e. it provides more information
and hence less variability, which reduces the frequency of
inference error occurrence
 Average run length
 An alternate measure of performance of control chart
 Denotes the no. of samples, on an average is required to
detect the out of control signal
 Let Pd denote the probability of observation plotting
outside the control limit, then run length is 1, with
probability of Pd and run length is 2, with probability of Pd(1
– Pd)
 ARL = Pd * ∑ j (1–Pd) j – 1 = 1 / Pd (Based on expansion of
infinite series)
18
Average Run Length – Contd..
 For a process in control, Pd = α
 For a 3σ control chart, detection of out of control
condition is given by ARL = 1 / Pd = 1 / 0.0026 = 385, i.e.
an observation will plot outside the control limit every
385 samples on average
 For a process in control, the ARL to be large, so that
observation plotting outside control limit is less or it does
not create a false alarm
 For an out of control process, the ARL should be small,
so that out of control can be easily detected. In this
case, Pd = 1 – β and hence ARL = 1 / 1 – β
 ARL curves can also be constructed for the control chart
based on the shift in mean with respect to ‘σ’ and ARL
 ARL can be expressed in terms of expected number of
individual unit sampled I. It is given by I = n (ARL)

19
Selection of Rational samples
 Shewhart described the fundamental criteria for
the selection of rational sub group
 Rational sample is chosen in such a manner that
the variations within sample is considered to be
due to common causes alone
 Samples are selected such that special causes if
present, will occur between samples
 i.e. the difference between sample will be
‘maximized’ while difference within sample is
‘minimized’
 It can be done by the instant-of-time method i.e.,
Observation are selected at approximately the
same time for the population under condition
20
 Selection of Rational samples – Contd..
 Sample size
 It is dependent on the expected degree of shift in the
process parameter
 For detecting slight change in the process parameter,
a large sample size is needed
 If we can tolerate smaller changes in the process
parameters, then a small sample size might suffice
 Frequency of Sampling
 It must be decided before constructing control charts

 Options are: choosing small sample size at frequent

intervals or choosing large sample size at infrequent
intervals
 Generally small sample size at frequent intervals is
adopted
21
 Selection of Rational samples –
Contd..
 Other factors that influence sample size
and frequency are
 Type of inspection for getting data –
destructive or non-destructive
 Current state of process
 Cost of sampling and inspection per unit
 Loss incurred due to non-conforming items
etc.

22
Analysis of patterns in control chart
 Criteria other than a plotted point falling outside
the control limits are also used to determine
whether a process is out of control
 Plot patterns often indicate whether the process
is in control or not
 Analysing these patterns is more difficult than
plotting the chart
 Identifying the causes of non-random patterns
require a knowledge of the process, equipment
and operation conditions as well as their impact
of characteristic of interest

23
Rules for Control chart

24
Analysis of control chart – contd..
 Rules
 A process is assumed to be out of control, if a single point
plots outside the control limits

25
Analysis of control chart – contd..
 Rules – contd.
 A process is assumed to be out of control, if two out of
three consecutive points fall outside the 2σ warning limits
on the same side of the centre line

26
Analysis of control chart – contd..
 Rules – contd.
 A process is assumed to be out of control, if four out of five
consecutive points fall beyond the 1σ unit on the same
side of the centre line

27
Analysis of control chart – contd..
 Rules – contd.
 A process is assumed to be out of control, if 9 or more
consecutive points fall to one side of the centre line

28
Analysis of control chart – contd..
 Rules – contd.
 A process is assumed to be out of control, if there is a run
of six or more consecutive points, steadily increasing or
decreasing

29
Analysis of control chart – contd..
 Rules – contd.
 For a process in control, roughly equal
number of points should be above or below
the centre line with no systematic patterns

30
 Interpretation of plots
 Five rules for determining out of control conditions are not
all used simultaneously
 Rule 1 is routinely used along with a couple of other rules
 Reason for not using all rules is that it increases the
chance of Type – I error, i.e., Probability of false alarm
increases
 Suppose a number of independent rules are used for out
of control criteria is ‘k’. Let αi be the probability of type I
error of rule ‘I’.
 Then the overall probability of type I error is α = 1 –
product (1 – αi)
 This is under the assumption that rules are independent,
but generally rules are not independent and hence it is
approximate only
 Even if all rules are applied, the process may be out of
control, for which the experience, judgement and
interpretive skills are needed
31
Other important considerations
 Determination of causes of out of control points
 Task of control chart is not just identification of out-of
control points, but should identify the causes behind that
 Requires a thorough knowledge of the process and also
the sensitivity of the output quality characteristic to
process parameter
 Determination of causes is a team effort
 Generally cause and effect diagram is used
 Maintenance of control chart
 Should be done on continuous basis, as it involves
changing the centre line and limits and when process goes
out of control
 Should be easily accessible to the person, who is
responsible for that quality characteristic
 Should be placed in shop floor so that all can know about
the current situation
32
 Example
• Chapter 6 – Problem No. 17, Page No. 258
• Suppose a 3-sigma control limits are
constructed for the average temperature in
furnace. Samples of size 4 were selected with
the average temperature of 5000 deg C and a
standard deviation of 50 deg. C.
– Find the 3-sigma control limits.
– Calculate probability of type I error for Rule 1
– Suppose Rule 2 and Rule 3 is used simultaneously to determine out of
control conditions. What is the overall probability of a Type I error,
assuming independence of the rules?
– Approximately how many samples, on average, will be analyzed before
detecting a change when Rules 2 and 3 are used simultaneously?
– What is the probability of the shift being detected within the first two
samples?

33
 Example – contd.
• Given
– Average temperature = mean = 5000 deg C
– Sample size (n) = 4
– Standard deviation is 50 deg C
• The centre line on a chart is equal to the
sample mean, which is 5000 deg C.
• The standard deviation of the sample mean
is

• Three sigma control limits are 5000 +/- 3(25)

= (4925, 5075)
34
 Example – contd.
• According to Rule 1, a process is said
to be out of control, if a point falls
outside the 3 control limits
• Using normal distribution the same
can be calculated
– Z1 = (4925 – 5000) / 25 = -3, From normal table, it
is 0.0013
– Z2 = (5075 – 5000) / 25 = 3, From normal table, it
is 0.9987. Hence the area above it is 1 – 0.9987 =
0.0013.
– Hence total probability of committing Type I error or
Probability of a point falling outside the control limit
is 0.0026
35
 Example – contd..
• Rules 2
– A process is assumed to be out of control, if two out of three
consecutive points fall outside the 2σ warning limits on the same
side of the centre line

36
 Example – contd.
• Rule 2 is probability of 2 out of 3 consecutive
points falling outside the two sigma limits, on a
one side of the centre line.
• 2 sigma limits can be calculated as follows:
– 5000 +/- 2(25) = (4950, 5050)
• Using normal distribution the probability of a point
falling outside the 2sigma limit can be calculated
as
– Z1 = (4950 – 5000) / 25 = -2, From normal table, it is 0.0228
– Z2 = (5050 – 5000) / 25 = 2, From normal table, it is 0.9772
Hence the area above it is 1 – 0.9772 = 0.0228.
• Probability of 2 out of 3 consecutive points falling
outside the two sigma limits, on a one side of the
centre line is . 3
C2 (0.0228) 2 (0.9772)  0.001524
• Since this can happen on either side, probability of
type I error using Rule 2 is 2 * 0.001524 = 0.003048.37
 Example – contd..
• Rule 3
– A process is assumed to be out of control, if four out of five
consecutive points fall beyond the 1σ unit on the same side of
the centre line

38
 Example – contd.
• Similarly, Rule 3 is probability of 4 out of 5
consecutive points falling outside the one sigma
limits, on a one side of the centre line is .
• 1 sigma limits can be calculated as follows:
– 5000 +/- 25 = (4975, 5025)
• Using normal distribution the probability of a point
falling outside the 1sigma limit can be calculated as
– Z1 = (4975 – 5000) / 25 = -1, From normal table, it is 0.1587
– Z2 = (5025 – 5000) / 25 = 1, From normal table, it is 0.8413
Hence the area above it is 1 – 0.8413 = 0.1587.
– Probability of 4 out of 5 consecutive points
falling outside the 1 sigma limits, on a one side
of the centre line is .
5
C4 (0.1587) 2 (0.8413)  0.002668
• Since this can happen on either side, probability of
type I error using Rule 3 is 2 * 0.002668 = 0.005336.
39
 Example – contd.
• Assuming independence of rules, the
probability of an overall type I error is
 1  (1  0.003048)(1  0.005336)  0.008368

• Average no. of samples required before

an out of control condition is indicated is
ARL = 1/0.008368 = 119.5 = 120

40
 Example – contd.
• Assuming the change in process mean, say
process average drops to 4960, the
standardized normal values at the control limits
are
– Z1 = 5075 – 4960 / 25 = 4.60
– Z2 = 4925 – 4960 / 25 = -1.40
• Using the standard normal tables, the area
above the UCL is 0.000 and that below the
LCL is 0.0808.
• The area between the control limits is 0.9192,
which is the probability of non detection of the
shift on a given subgroup.
41
Example – Contd..

42
 Example – contd.
• Now the probability of detecting shift on the first
subgroup is 0.0808.
• Next the probability of not detecting the shift on the
first subgroup and detecting on the second sub group
is 0.9192 * 0.0808 = 0.0743.
• Similarly the probability of not detecting the shift on the
first two subgroups and detecting on the third
subgroup is (0.9192)(0.9192)(0.0808) = 0.0683,
assuming independence of subgroups.
• Hence the probability of detecting the change by the
third subgroup is 0.0808 + 0.0743 + 0.0683 = 0.2234.
• Thus, the probability of failing to detect the change by
the third subgroup point drawn after the change is 1 –
0.2234 = 0.7766,
• Using the computation above, The probability of the
shift being detecting within the first two subgroups is
0.0808 + 0.0743 = 0.1551 43