Statistical Quality Control

BITS Pilani Dr. Sudeep Kumar Pradhan, PhD.

Pilani Campus
 BITS Pilani
Pilani Campus

Control Charts for Variables

2
Introduction

• Variables are those quality characteristics that are measurable

on a numerical scale, e.g. length, viscosity
• It is necessary to control the mean value of a quality
characteristic as well as its variability
• Mean gives an indication of the central tendency of a process
• Variability provides an idea of the process dispersion
• When there is a change in process mean or process standard
deviation, the proportion of parts that do not meet specification
increases
• Control chart aid in detecting such changes in process parameters
• Variables provide more information than attributes
• Attributes deal with qualitative information, whether an item is
non-conforming or not ok, but didn’t show the degree to which a
quality characteristic is non conforming
• Eg. 40 ± 0.5. Both 40.6 and 42 are nonconforming

3
BITS Pilani, Pilani Campus
Introduction – Contd..

• Cost of obtaining variable data is usually higher than for

attributes because
– Attribute data is collected by means such as go / no-go gages,
which are easier to use and hence less costly also
– Total cost of data collection is the sum of two costs – fixed and
variable cost
– Fixed cost include cost of inspection equipment, while variable
cost include cost of inspecting
– More the parts inspected, higher the variable costs and fixed
cost is unaffected
– With the use of automated devices for measurement, the
differences in variable unit cost between variable and attribute
may not be much, but fixed cost may increase
– Attribute chart for the proportion of non-conforming may
represent the general operational level of the plant but not the
variable chart
4
BITS Pilani, Pilani Campus
Selection of Characteristics of
Investigation
• A single component has usually several quality
characteristic
• Number of quality characteristic that affect a product
will be large
• Not feasible to maintain control chart for all as decision-
making becomes difficult
• Involves selecting a few vital characteristics and
selection is based on those that cause more non-
conforming items and increase cost
• Pareto analysis plays a major role in selection
• It is desirable to measure process characteristic that
have a casual relationship to product characteristic

5
BITS Pilani, Pilani Campus
Preliminary decisions

• Selection of Rational samples

– Sampling method should maximize differences
between samples and minimize differences within
sample
– Lot from which samples are chosen should be
homogeneous
– For detecting shift in process parameter, samples
should be made up of item produced at same time
– For non-conformance of items produced since the
previous sample, then samples should be selected
from items produced since that time

6
BITS Pilani, Pilani Campus
Preliminary decisions – contd..

• Sample size
– Size is normally between 4 and 10 and in industry it
will be 4 to 5
– Larger the sample size, better the chance of detecting
small shifts
– Based on factors like cost of inspection or cost of
shipping a nonconforming item to the customer etc.
• Frequency of sampling
– Depends on the cost of obtaining information
compared to the cost of non detecting a non
conforming item
– As process is brought to control, frequency of
sampling decreases

7
BITS Pilani, Pilani Campus
Preliminary decisions – Contd..

• Choice of measuring instruments

– Accuracy of the instruments directly influences the
quality of the data calculated
– Should be calibrated and tested
– Characteristic being controlled, the desired degree of
measurement precision, both impacts the selection
• Design of data recording forms
– Recording forms should be designed based on the
control chart to be used
– Includes sample no., date, time, raw values of data, part
name, lot number, operator name, machine, gauge, unit
of measurement, specification etc.

8
BITS Pilani, Pilani Campus
Control chart format

9
BITS Pilani, Pilani Campus
Control chart for Mean and Range

• Development of chart
– Using a pre-selected scheme and sample size record
measurements of the selected quality characteristic
– For each sample, calculate the sample mean and range
Ri = xmax - xmin x1  x 2    x n
x
n
– Obtain and draw the centre line and trial control limit
• Find the mean of all sample mean (Formula)
• Find the mean range of all samples (Formula)
– 3σ control limits for Mean chart is given by
 ˆ
 Z / 2 x  Z / 2 X  3 X  X  3
n n
– For normally distributed population, the distribution of the
statistic’s relative range (w) = R / σ and it is dependent on
sample size ‘n’
10
BITS Pilani, Pilani Campus
Control Chart for Mean (Xbar)

11
BITS Pilani, Pilani Campus
Control chart for Mean and Range –
Contd..
– Mean of w is given by d2
– Estimate of the process standard deviation is
ˆ R /d 2
R
– (UCL, LCL) = X 3  X A2 R
d2 . n
R 3 R
– Control limits for R chart is given by
– R = σ.w and hence σR = σ. σw
– Mean of σ
wˆ is
Rgiven
/d 2 as d3 and we Rknow that
also
 R  .d 3
Hence d2
D3 R D4 R
– UCL = LCL =
12
BITS Pilani, Pilani Campus
Control Chart for Range (R)

13
BITS Pilani, Pilani Campus
Control chart for Mean and Range –
Contd..
• Plot the values of the range on the chart and find
whether points are in statistical control
– An R chart is analysed before X-bar chart to determine
out of control situations, as R chart reflect process
variability, which should be brought to control.
– If R chart shows out of control, then the X-bar chart is
meaningless
• Delete the out of control points for which remedial
action has been taken to remove special causes and
the remaining samples are used to obtain revised
limits
• A point of interest is about the point that falls
below the LCL, when LCL is greater than zero
14
BITS Pilani, Pilani Campus
Control chart for Mean and Range –
Contd..
• These points are desirable because they indicate
unusually small variability, within the sample
and might be due to special causes
• This condition helps in further reducing our
process variability
• Implement the control chart
Why two charts?
• X bar chart monitors the mean between sample
values
• R chart monitors the variation within sample.

15
BITS Pilani, Pilani Campus
Example

16
BITS Pilani, Pilani Campus
Example – contd..

17
BITS Pilani, Pilani Campus
Example – contd..

18
BITS Pilani, Pilani Campus
Example – contd..

In this case, D3 = 0 for n = 5

and hence LCL = 0
19
BITS Pilani, Pilani Campus
Example – contd..

20
BITS Pilani, Pilani Campus
Example – contd..

21
BITS Pilani, Pilani Campus
22
BITS Pilani, Pilani Campus
Standardized control chart

• When the sample size varies, it results in fluctuating

control limits and hence we need standardized control
chart
• Standardized values represent a deviation from the
mean in units of standard deviation
• They are dimensionless and have a mean of zero
• Control limits for this are ±3 and are constant
• Easier to interpret the shift in the process
• Sample size for sample i is ni and let Xi* and si denotes
g
its mean and standard deviation
 ni X i
• Mean of sample average X  i 1g
n i
23 i 1
BITS Pilani, Pilani Campus
Standardized control chart – Contd..

• An estimate of the process standard deviation σ* is

given by g

 (n
2
i  1) si
ˆ  i 1
g

 (n
i 1
i  1)
• For sample i, standardized value for the mean zi is
Xi  Xi
zi 
ˆ
ni
• A plot of the Zi values on a control chart with the centre
line at ‘0’ , UCL at 3 and LCL at -3 is standardized X-bar
chart

24
BITS Pilani, Pilani Campus
Standardized control chart – Contd..

– To standardize a Range chart, the range Ri for sample ‘i’

is first divided by the estimate
Ri of the process standard
deviation ri 
ˆ
ri  d 2
ki 
– Standardized value of Range is given by d3

– Value of ki are plotted with a centre line at ‘0’ and UCL =

‘3’ and LCL = ‘-3’
• Control Limits for a given target or Standard
– Management may specify values for the process mean or
standard deviation
– These values are the goals or desirable standards or
target values, which helps to determine process able to
meet standard 25
BITS Pilani, Pilani Campus
Control limits for a given target or
Standard - Contd..
– Let X o and  o represent the target values of the
process mean and standard deviation respectively
and the centre line and limits are given by
CL X X 0
0
UCLX X 0  3 .  X 0  A . 0
n
0
LCL X X 0  3 .  X 0  A . 0
n
– For R - chart
R
CLR d 2 . 0 (ˆ  )
d2
UCLR R  3 . R  d 2 . 0  3 . d 3 0 D2 . 0
LCLR R  3 . R  d 2 . 0  3 . d 3 0 D1. 0
26
BITS Pilani, Pilani Campus
Interpretation and Inferences

• Caution
– Sample plots may fall outside the limit, even
though no special causes are present
– Reason being that desirable standards may
not be consistent with the process conditions
– It is easy to meet a desirable target value for
process mean than it is for process
variability (Range)

27
BITS Pilani, Pilani Campus
Interpretation and Inferences –
Contd..
• Interpretation
– Difficult and needs thorough knowledge
about different process parameters on
quality characteristic
– When R-chart is brought to control, many
special causes for the Xbar chart are
eliminated as well
– Xbar chart monitors the centering of the
process and a jump indicates process
average has jumped
28
BITS Pilani, Pilani Campus
Interpretation and Inferences – Contd..

– Increasing trend indicates the process centre is

gradually increasing and requires adjustments in
machine settings or such controllable parameters as
proper tool, proper depth of cut, feed etc.
– Reducing process variability in order to allow an R-
chart to exhibit control is difficult and needs “quality
improvement”
– Process capability can be estimated from the process
standard deviation, and helps to determine how the
process performs with respect to specification limits
and also in determining the proportion of output that
is non conforming

29
BITS Pilani, Pilani Campus
Control chart patterns and corrective
action
• A ‘non-random identifiable’ pattern in the
plot of a chart might provide reason to look
for special cause in a process
• There are about 15 typical patterns
identified by Western Electric company and
9 of them have been discussed here
• Natural Patterns
– No identifiable arrangement of the plotted point
exists
– No point fall outside the control limits
– Majority of the points are near the centre line and
few points close to control limits
– Demonstrates the presence of stable system of
common causes 30
BITS Pilani, Pilani Campus
Control chart patterns – contd..

31
BITS Pilani, Pilani Campus
Control chart patterns – Contd.

• Sudden shifts in the level

– Can occur because of changes intentional or
otherwise in process settings (temperature, depth of
cut etc.)
– New operators, new equipment, new vendors, new
methods are the reasons for sudden shift

32
BITS Pilani, Pilani Campus
Control chart patterns– Contd..

• Gradual shifts in the level

– Occurs when a process parameter changes gradually over a period
of time
– X-bar chart might exhibit such a shift because of incoming quality
of raw materials that would have changed with time
– Change in style of supervision, maintenance program etc.
– R-chart might exhibit such a shift because of new operator,
decrease in worker skill due to fatigue etc.

33
BITS Pilani, Pilani Campus
Control chart patterns – Contd..

• Trending pattern
– Differs from gradual shift in level, that trends do not stabilize or settle
down
– Represents changes that steadily increase or decrease
– For X-bar chart, can be due to tool wear, deterioration of equipment,
build up of debris on jigs and fixtures, change in temperature etc.
– For R-chart, it may be due to improvement in operator skill due to on
job training, decrease due to fatigue etc.

34
BITS Pilani, Pilani Campus
Control chart patterns – Contd..

• Cyclic patterns
– Characterized by repetitive periodic behaviour in the system
– Cycles of low and high points will appear on the control chart
– X-bar chart may exhibit because of rotation of operators, periodic change of
temperature or humidity, seasonal variation in incoming components
– R-chart may exhibit this pattern because of operator fatigue and getting
energized in subsequent breaks, a difference between shifts, periodic
maintenance of equipments etc.
– If samples are taken so infrequently, only the high and low points will be
represented

35
BITS Pilani, Pilani Campus
Control chart patterns – Contd..

• Wild patterns
– Can be classified as Bunches and Freaks
– Cluster of several observation that are different
from other points and special causes are associated
with these points
– Freaks
• are caused by external disturbances that influence one or
more samples
• They are points that are too small or large with respect to
control limits and fall outside the control limits and
hence easy to identify
• Care should be taken that no measurement or recording
error is associated with that freak point
• Some special causes may be sudden, very short-lived
power failures, use of new tool for a brief test period etc.

36
BITS Pilani, Pilani Campus
Control chart patterns – Contd..

37
BITS Pilani, Pilani Campus
Control chart patterns and corrective
action – Contd..
• Wild patterns – Contd.
– Bunches
• Cluster of several observation that are different from other
points
• Possible causes may be use of new vendor, use of a different
machine, use of new operator etc., for a short time period.

38
BITS Pilani, Pilani Campus
Control chart patterns and corrective
action – Contd..
• Mixture patterns
– Effect of two or more population in the sample
– Characterized by points that fall near the control limits, with
absence of points near the centre line
– Might be due to material from two different vendors, different
production method, two or more machine being represented

39
BITS Pilani, Pilani Campus
Control chart patterns and corrective
action – Contd..
• Stratification patterns
– Is also due to presence of two or more population distribution
– Output is combined or mixed and samples are selected from it
– Majority of the points fall close to centre line, with very few points
near the control limits
– Can be misinterpreted as indicating unusually good control

40
BITS Pilani, Pilani Campus
Control chart patterns and corrective
action – Contd..
• Interaction patterns
– Occurs when the level of one variable affects the
behaviour of other variables associated with the quality
characteristic
– Interaction pattern can be detected by changing the
scheme for rational sampling
– Example, low pressure and low temperature may
produce a desirable effect on output characteristic
– Effective sampling method would involve controlling
the temperature at several high values and then
determining the effect of pressure on output
characteristic for each temperature value
– If the R-chart shows the sample range to be small, then
information regarding the interaction could be used to
establish desirable process parameter settings

41
BITS Pilani, Pilani Campus
Control chart patterns – Contd..

• Interaction patterns

42
BITS Pilani, Pilani Campus
43
BITS Pilani, Pilani Campus
Control chart for Mean and Standard
deviation (X-bar and s chart)

• R-chart is easy to construct and use, but a

standard deviation chart (s-chart) is preferable
for large sample sizes (> 10)
• Reason is that range accounts only for the
maximum and minimum sample values and is
less effective for large samples, where as sample
standard deviation serves as a better measure of
process variability
• Population distribution of a quality
characteristic is assumed to be normal with a
population standard deviation denoted by σ
44
BITS Pilani, Pilani Campus
45
BITS Pilani, Pilani Campus
Control chart for Mean and Standard
deviation (X-bar and s chart)

• Mean and standard deviation of the sample

standard deviation is given by
E (s) c4
 s  . 1  c4 2 where c4
depends on the sample size given by

1  n  2   !
 2    2 
2 
c4    n  3  
 n  1     !
  2  
46
BITS Pilani, Pilani Campus
Control chart for Mean and Standard
deviation (X-bar and s chart) – Contd..
g
• No Given Standards s i
– Centre line of a s-chart is CLs  s  i 1
g
s
where g is the no. of samples and ˆ  c
4
si is the standard deviation of the ith sample
– UCL s = s  3 .  s  s  3 .  . 1  c 4
2

– Estimate of a population standard deviation is

– Hence control limits are given by
s 2
UCLs  s  3. . 1  c4  B4 s
c4
s 2
LCLs  s  3. . 1  c4  B3 s
c4 47
BITS Pilani, Pilani Campus
Control chart for Mean and Standard
deviation (X-bar and s chart) – Contd..

• No Given Standards
• X-bar chart
– Centre line is similar to that of X-bar chart
– Control limits are given by
s
UCLX  X  3.  X  A3 s
c4 n
s
LCLX  X  3.  X  A3 s
c4 n
• S-chart is constructed first as the standard deviation of
X-bar is dependent on ‘s’ and if the s-chart is not in
control, any estimate of the standard deviation of X-bar
chart is unreliable
48
BITS Pilani, Pilani Campus
Control chart for Mean and Standard
deviation (X-bar and s chart) – Contd..

• Given Standard
– If the target standard deviation is given as σ0 then, centre line of
a s-chart is CLs  c4 0
– Hence control limits are given by
2
UCLs  c4 0  3 s  c4 0  3. 0 . 1  c4  B6 0
2
LCLs  c4 0  3 s  c4 0  3. 0 . 1  c4  B5 0
– X-bar Chart
• Target value is specified as Xo then control limits are given
by CL X X 0

3
UCLX  X 0  A 0 , where A 
n
LCLX  X 0  A 0 49
BITS Pilani, Pilani Campus
Example

The thickness of the magnetic coating on audio tapes is an

important
characteristic. Random samples of size 4 are selected, and the
thickness is measured using an optical instrument. Table 7-3 shows
the mean Xand standard deviation s for 20 samples. The
specifications are 38 ±4.5 micrometers. If a coating thickness is less
than the specifications call for, that tape can be used for a different
purpose by running it through another coating operation.(Sum
X=743.5, Sum s=95.8)

50
BITS Pilani, Pilani Campus
(a) Find the trial control limits for an X- and an s-chart.

(b)Assuming special causes for the out-of-control points,

determine the revised control limits.

(c) Assuming the thickness of the coating to be normally

distributed, what proportion of the product will not meet
specifications?

(d)Comment on the ability of the process to produce items

that meet specifications.

(e) If the process average shifts to 37.8 micrometre, what

proportion of the product will be acceptable?

51
BITS Pilani, Pilani Campus
52
BITS Pilani, Pilani Campus
53
BITS Pilani, Pilani Campus
54
BITS Pilani, Pilani Campus
Thank You

55
BITS Pilani, Pilani Campus