SQC& OR UNIT, PUNE

Course Material of
Six Sigma Black-belt

- Measure Phase
 TABLE OF CONTENT
SQC& OR UNIT, PUNE

3 Measure
3.1 Data Collection Plan- CTQ tree
3.2 Measurement Systems Analysis-
Concepts
3.3 Baselining
3.4 Process Capability Analysis
including Sigma Level Estimation
 Overview of Measure Phase
SQC& OR UNIT, PUNE

Measure: Ensure that the measurement uncertainty of

CTQ's data are known and suitable and establish baseline
process capabilities.
1. Prepare Data Collection Plan
• CTQ-tree augmented with QC Support System

2. Measurement Systems Analysis

• CTQ-tree augmented with Measurement Support
System
• Accuracy
• Bias
• Precision
 Overview of Measure Phase
SQC& OR UNIT, PUNE

3. Base-lining
• Type of Data
• Time Series Plot
• Histogram
• Frequency Plot
• Distributions- Normal, Binomial & Poisson
• Normality test
• Sampling
 Overview of Measure Phase
SQC& OR UNIT, PUNE

4. Process Capability Analysis including Sigma

Level Estimation
• Mean-Median-Mode
• Dispersion- Range, Standard Deviation
• Cp, Cpk
• Pp, Ppk
• Sigma Level- Short-term and Long-term
SQC& OR UNIT, PUNE

CHAPTER – 3.1
Data Collection Plan
- CTQ Tree
 DATA COLLECTION PLAN
SQC& OR UNIT, PUNE

Most important aspect of the Measure Phase is to

collect data of CTQs and drill-down parameters of
CTQs. Need is to prepare a Data Collection Plan
which would be an augmentation in the CTQ Tree
itself. The Data Collection Plan includes:
• Type of Data
• Unit of Measures
• Minimum Frequency of Data Collection
• Specifications/ Requirements
• Responsibility
• Reaction Plan in case of Sporadic Problems
• Current Controls in the form of SOPs for the
CTQ
 TYPE OF DATA
SQC& OR UNIT, PUNE

Two types of Quantitative Characteristics are

available:

Attribute/Discrete: The characteristics which assume

only isolated values in its range of variation is
termed as Discrete Characteristics. e.g. No. of
complaints per month, Percentage absenteeism, No.
of injuries etc.

Variable/ Continuous: The characteristics which

may assume any value within its range of variation
are termed as Variable or Continuous
Characteristics. e.g. Height, Weight, Diameter etc.
 TYPE OF SPECIFICATIONS
SQC& OR UNIT, PUNE

Specifications are representation of voice of the

customers.
Quantitative characteristics are specified as:
Target is the best: Example:Hb level of Male”
Upper specification Limit (USL): 16
Target : 14
Lower Specification Limit (LSL): 12

Higher the Better: LSL: e.g. Yield: >= 95%.

Lower The Better: USL: e.g. # %ge Reject <=5%.

 DATA COLLECTION PLAN IN CTQ TREE
SQC& OR UNIT, PUNE

Data collection Plan is augmented in the CTQ tree

itself as under:
Code Z1 Z2 Z3 Z4 Y X Y1
CTQs OTD% # wrong Revenue Profit Cycle time No. of P1-CT
delivery employees
Type of Data Variable Discrete Variable Variable Variable Discrete Variable
UoM % No. Rs. Rs. Sec
USL 60
Target
LSL 90
Min freq. of Each Quarterly Each transaction
data: transactio
n
Record ref. CTQ tree
Responsibility GB
SQC& OR UNIT, PUNE

CHAPTER – 3.2
Measurement System Analysis
 MEASUREMENT SYSTEM PLANNING
SQC& OR UNIT, PUNE

Measurement System Planning is incorporated as

Measurement System Support Augmentation in the
CTQ Tree as under:
Code Z1 Z2 Z3 Z4 Y X Y1
CTQs OTD% # wrong Revenue Profit Cycle time No. of P1-CT
delivery employees
Type of Data Variable Discrete Variable Variable Variable Discrete Variable
UoM % No. Rs. Rs. Sec
Accuracy: Instrument: Clock of GM
Make
L.C.
Calibration SOP…
Bias
Precision: %R&R 0.79
Sigma R&R 1.1
 MEASUREMENT SYSTEM PLANNING
SQC& OR UNIT, PUNE

Measurement System Planning is incorporated

immediately below the QC System Support
Augmentation in the CTQ Tree. This Planning
includes:
- Instrument/ Measurement Device
- Selection (Make, Manual, Least Count etc.)
- Calibration details
- Maintenance of the instrument
- Conditioning/ handling/ preservation of the
instruments
- Accuracy &
- Precision
 WHAT IS MSA?
SQC& OR UNIT, PUNE

 The process of measurement involves variation.

Any variation in measurement that is taken is
made up of the true value and the error.
 The error is made up of variation inherent in the
measuring equipment, and the individual or
method used in measuring.
 MSA tries to understand, measure and analyze
the error.
 WHAT IS MSA?
SQC& OR UNIT, PUNE

Variation in a process
(CTQ)

Due to Object Due to Measurement

(Between Parts / Process
Within a part)

Accuracy Precision
 COMPONENTS OF MSA
SQC& OR UNIT, PUNE

Accuracy Precision

Difference between actual Variation in repeat

(true) and measured values measurements

 Accurate  Accurate  Accurate  Accurate

 Precise  Precise  Precise  Precise
 BIAS
SQC& OR UNIT, PUNE

“How accurate is the Gauge when compared to a master value?”

A Gauge accuracy study examines the

difference between the observed average
measurement and a reference or master value.
True or Reference
Value

Bias

Observed
Average value
 ACCURACY
SQC& OR UNIT, PUNE

Difference between actual

(true) and measured values

Linearity Bias Stability

Size vs Accuracy Average vs Master Time vs Accuracy

Difference in the Difference between Total variation
observed accuracy
the observed average observed over time on
values through the
expected range of measurement and a a CTQ. Keeping all
measurements master (true) value. other constant.

Gauge Linearity & Accuracy Study

 PRECISION
SQC& OR UNIT, PUNE

Variation in repeat
measurements

Repeatability Reproducibility

Variation due to the Variation due to

measuring device. measurement system.
Variation observed
Different operators
when same operator
measures same part measure the same
repeatedly with the object using the same
same device. device.

Gauge R & R Studies.

 EXAMPLE
SQC& OR UNIT, PUNE

Three batches were selected that represent the expected range

of the process variation. Three inspectors measured the three
batches, three times per batch, in a random order for a CTQ
called selectivity.

Inspector A B C
Batches 1 2 3 1 2 3 1 2 3
1 4.05 3.89 3.84 2.74 1.58 3.74 4.76 4.56 4.46

2 4.09 4.06 3.86 6.09 5.31 4.79 4.44 4.35 4.63

3 3.69 4.26 4.14 2.70 4.71 2.69 3.99 4.33 4.20

Data is presented in a coded form

SQC& OR UNIT, PUNE

Gage name:
Date of study:
Gage R&R (Xbar/R) for Selectivity Reported by:
Tolerance:
Misc:

Xbar Chart by Inspector Gage name: Inspector * Objects Interaction

1 2 Gage R&R3 (Xbar/R) for Selectivity
Date of study:
4.9
Reported by :
Inspector
5.5 3.0SL=5.557 Toleranc e:
Misc : 1
Sample Mean

Xbar Chart by Operator Operator*Part Interaction 2

4.4 Operator
4.5 4.9

Average
1 2 3
3.0SL=5.557
5.5
1 3
Sample Mean
X=4.062 4.4 2

Average
4.5
3

3.5 3.5
X=4.062
3.9
3.9
2.5 -3.0SL=2.567 3.4
Part
2.5 -3.0SL=2.567 3.4
1 2 3

R Chart by Operator Response by Operator

4 1 2 3
3.0SL=3.761 6
Obj. 1 2 3
Sample Range

3 5

2 4

R Chart by Inspector 1
R=1.461
3 Response by Inspector
2
4 1 2 0
3 -3.0SL=0.00E+00
6
3.0SL=3.761 Operator 1 2 3

Components of Variation Response by Part

3 5
Sample Range

100 6
% Total Var
% Study Var 5
4
Percent

2 50 4

R=1.461 3

1 2 3
0
Gage R&R Repeat Reprod Part-to-Part Part 1 2 3

0 -3.0SL=0.00E+00 2

Inspector 1 2 3

Components of Variation Response by Objects

100 6
%Total Var
%Study Var 5
Percent

4
50
3

2
0
Gage R&R Repeat Reprod Obj. to Obj. Obj. 1 2 3
SQC& OR UNIT, PUNE

Gage R&R for Selectivity

Source Variance Std.Dev. 5.15*Sigma
Total Gage R&R 0.721622 0.849483 4.37484
Repeatability 0.721622 0.849483 4.37484
Reproducibility 0.000000 0.000000 0.00000
Part-to-Part 0.202212 0.449680 2.31585
Total Variation 0.923834 0.961163 4.94999

Source %Contribution %Study Var

Total Gage R&R 78.112 88.381
Repeatability 78.112 88.381
Reproducibility 0.000 0.000
Part-to-Part 21.888 46.785
Total Variation 100.000 100.000
SQC& OR UNIT, PUNE

Gage R&R (Continuous Data)

Gage R&R Results
1). Two-Way ANOVA Significant?

Part p-value Y N
Oper p-value Y N
Oper & Part p-value Y N

Pass?

2). % Tolerance [ <30% ] Y N

3). % Contribution [ <10% ] Y N
4). % Study [ <30% ] Y N
5). # Distinct Categories [ >=4% ] Y N

Graphical Output OK?

1). Effective Resolution [ >50% ] Y N

2). Stability [R Chart] Y N
3). Consistency Between Xbar consistency Y N
Between Oper
4). Systematic Shift [Oper/Part Inter. Plot] Y N

Gage R&R Pass? Y N, If NO

Plan for improvement:
 MSA FOR ATTRIBUTE DATA
SQC& OR UNIT, PUNE

This section focuses on MSA for data that are nominal or

ordinal.
Nominal data classifies occurrences into unordered categories,
for example:
 A sales request can or cannot be filled
 A customer survey response on a satisfaction
questionnaire is either “yes” or “no”
 A customer contact is considered acceptable or not
 The origin of a sales contact is recorded (can be one of
several origins)
 Categories of idle time
 MSA FOR ATTRIBUTE DATA
SQC& OR UNIT, PUNE

Ordinal data classifies occurrences into ordered categories, for

example:
 A customer survey response on a questionnaire is
“strongly disagree”, “moderately disagree”, “neutral”,
“moderately agree”, or “strongly agree”
 A sales lead is considered “unlikely”, “likely”, or “very
promising”
 An IT failure is classified as being of “low”, “medium”,
or “high” severity
 Categories of job difficulty
 ASSESSING BIAS
SQC& OR UNIT, PUNE

Example: Consider an example

dealing with security violations.
Two Auditors and one Expert Auditor
inspected 25 areas chosen from
many areas that could have been
inspected, and classified each as P
or F (pass or fail).
An attempt was made to include
areas with security violations as
well as without (there should be at
least 25% representation of each
group).
 ASSESSING BIAS
SQC& OR UNIT, PUNE

Auditor 1 agrees with the Expert for 24 of the 25 areas.

Auditor 2 agrees with the Expert for 18 of the 25 areas.
The percent agreement is dependent upon the areas chosen; had
different areas been chosen, the sample percent agreement
would differ.
In other words, the percent agreement is a sample statistic and
is subject to sampling variation.
The sample percent agreement is an estimate of a population
parameter, the true percent agreement.
Thus it is useful to calculate a range of values that is likely to
contain the true percent agreement with some high probability.
SQC& OR UNIT, PUNE

CHAPTER – 3.3
Baselining
 BASELINING
SQC& OR UNIT, PUNE

• Baseline status represents the quality status of a CTQ of

the present status.
• Present status may or may not be predictable (stable); but
preference of the present status would be to have a
predictable status.
• Baseline status may or may not be normal with respect to
the distribution.
• Predictable best performance of the present process should
be represented as the baseline status.
• Past data of CTQ are collected, examined for trend to
select baseline data, selected data is examined with respect
to histogram, test of statistical distributions, control chart to
examine to isolate sporadic problems and finally represent
the data belonging to the predictable baseline data.
 TIME SERIES PLOT- Data
SQC& OR UNIT, PUNE

• Performance in terms of Specific electricity

Consumption Index (Unit consumed / Production volume
per month) over month are collected and appearing in
Minitab as under:
 TIME SERIES PLOT
SQC& OR UNIT, PUNE

• Performance over time graphically plot is the time

series plot.
• Data of CTQs are collected in the progress of time and
plotted in the Time Series Plot in Minitab-13 as under:
 TIME SERIES PLOT
SQC& OR UNIT, PUNE

• Preparatory to the plot in minitab is done as under:

 TIME SERIES PLOT
SQC& OR UNIT, PUNE

• Time Series Plot is available as under:

• This portion should be selected as baseline.

 HISTOGRAM
SQC& OR UNIT, PUNE

• Selected data should be plotted for histogram to look

for the nature of distributions.

• Number of data points should be at least 50 to construct

a distribution. Selected Specific Electricity Consumption
Index data are less in no. So, the histogram is not able
show any regular shape.
 HISTOGRAM
SQC& OR UNIT, PUNE

• Normally one will look for testing the data for normal
distribution of variable data.
Normality Test in Minitab
SQC& OR UNIT, PUNE
 Normality Test in Minitab
SQC& OR UNIT, PUNE

Data Normal if
P>0.05
 Normality Test in Minitab
SQC& OR UNIT, PUNE

• In the given example, for the select data:

Distribution
is Normal
as P>0.05
 Probability Distributions
SQC& OR UNIT, PUNE

Need quantify/verify our conclusions from the descriptive

statistics investigation:

•Introduce probability distributions for use in analytical statistical methods

•Remove the subjectivity from the use of descriptive statistics investigations

The normal distribution is the basis of the ANOVA method

 Probability Distributions: Terminology
SQC& OR UNIT, PUNE

Given a Random Variable:

• Probability Density Function (PDF)

– The Frequency of occurrence for the variable.

• Cumulative Distribution Function (CDF)

– The Probability that the variable is less than or equal to some value
– It is the area under the PDF and ranges from 0 to 1.

• Reliability Function
– The Probability that the variable is greater than some value
– It is 1-CDF
 Continuous Distribution: Normal
SQC& OR UNIT, PUNE

Description
• The normal distribution (also called the Gaussian distribution) is the
most commonly used distribution in statistics. Two parameters (m (mu)
and s (sigma)) are required to specify the distribution.

The distribution
 x  m 2

x ; m  1 
p ,s 2
e 2s 2

2 s 2

Notes
• The normal distribution closely matches the distribution of many
random processes, especially measuring processes

• Much of what is done in MAIC and DFSS is based on the normal

distribution
 Normal CDF & Reliability Function
SQC& OR UNIT, PUNE

1.0

P(x < X)
0.5

0.0
-4 -3 -2 -1 0 1 2 3 4
x
Cumulative
Distribution
Function (CDF)

1.0
P(x > X)

0.5

0.0
-4 -3 -2 -1 0 1 2 3 4
x
Reliability
Function
 Parameters of the Normal Distribution
SQC& OR UNIT, PUNE

m (mu), a measure of central tendency, is the mean or average of all values in

the population. When only a sample of the population is being described, mean
is more properly denoted by x (x bar).

s (sigma) is a measure of dispersion or variability. With smaller values of s, all

values in the population lie closer to the mean. When only a sample of the
population is being described, the standard deviation is more properly denoted
by s.

Both m and s are specific values for any given population, and they change as
the members of the population (the distribution) vary.
 A Plot of the Normal Distribution
SQC& OR UNIT, PUNE

mean or average ( m or x )

standard deviation (s)

probability of USL defect

LSL USL
lower specification limit upper specification limit
SQC& OR UNIT, PUNE
SQC& OR UNIT, PUNE
SQC& OR UNIT, PUNE
 Formal Definitions of Moments:
Statistical Expectation
SQC& OR UNIT, PUNE

The mean
• also called Expected Value or First Moment
• the Mean is a measure of central tendency, i.e., “Where is the center of
the distribution?” m  E  X 

 
i1
x i f ( xi) for discrete variables

  
x f ( x ) d x fo r c o n tin u o u s v a ria b le s

Variance
• also called Second Moment
• Variance is a measure of spread in the distribution
2
m 2  s  Var  X   E  X  m 2 

   xi  m 2 f ( xi ) fo r d is c re te v a ria b le s
i  1



 
m  x 2 f ( x ) d x fo r c o n tin u o u s v a ria b le s
 Higher Order Moments
SQC& OR UNIT, PUNE

Third order moment

• the third order moment is a measure of assymetry in the distribution
• when normalized, a skewness value is computed
• Skewness is zero for symmetric distributions
m 3  E   X  m 3 

m
skew ness    3
s 3

Fourth order moment

• the fourth order moment is a measure of the tail-heaviness of the
distribution
• when normalized, a kurtosis value is computed
m 4  E   X  m 4 

m
kurtosis    4
s 4
 Moments for Distributions
SQC& OR UNIT, PUNE

d is tr ib u tio n m ean v a r ia n c e skew n ess k u r to s is

b in o m ia l n p n p 1  p  1  2 p 1  6 p 1  p 
3 
n p 1  p  n p 1  p 

P o is s o n  t  t 1 1  3 t
t t
m s 2
n o rm a l 0 3

u n ifo r m
xU SL  x L SL x U SL  x L SL  2
0 1 .8
2
12
2 
3   4 
c h i- s q u a r e d  8



Only if we are experiencing unusually high defect rates for a

given mean and variance do we worry about skewness and
kurtosis.
 NORMAL DISTRIBUTION
SQC& OR UNIT, PUNE

Collect more data and understand variation

92 86 84 83 93 80 96 85 89 84
86 70 80 88 83 84 84 89 81 90
99 80 88 83 89 99 86 80 98 72
80 88 86 93 93 97 85 84 87 82
92 83 73 78 99 99 91 86 89 94
86 85 85 97 85 90 83 83 88 86
88 85 85 92 87 79 81 82 76 86
92 77 85 78 77 80 72 83 83 70
84 84 86 83 93 81 87 85 74 78
91 89 98 89 95 89 97 77 95 76

Very difficult to understand and comprehend

How can we understand?
 UNDERSTANDING VARIATION
SQC& OR UNIT, PUNE

Collect more data and understand variation

92 86 84 83 93 80 96 85 89 84
86 70 80 88 83 84 84 89 81 90
99 80 88 83 89 99 86 80 98 72
80 88 86 93 93 97 85 84 87 82
92 83 73 78 99 99 91 86 89 94
86 85 85 97 85 90 83 83 88 86
88 85 85 92 87 79 81 82 76 86
92 77 85 78 77 80 72 83 83 70
84 84 86 83 93 81 87 85 74 78
91 89 98 89 95 89 97 77 95 76

Very difficult to understand and comprehend

How can we understand?
 UNDERSTANDING VARIATION
SQC& OR UNIT, PUNE

Post the data to classes such as 65-70, 70-75, …, 95-100

and construct a frequency diagram

A pattern emerges. What is so special of this pattern?

 CHARACTERIZING THE PATTERN
SQC& OR UNIT, PUNE

Uni-modal, Symmetric from centre, Centre being average,

median, and mode
Normal Distribution

What about variability?

 CHARACTERIZING THE PATTERN
SQC& OR UNIT, PUNE

Variability is estimated through standard deviation.

Variation /
Standard deviation
s
Mean or Average
m

Center / Location

How do you calculate standard deviation?

 DESCRIPTIVE STATISTICS
SQC& OR UNIT, PUNE

Average or Mean - Sum of all Data divided by number of data points

Median - Middle Data Value when data is ranked from min. to max.

Mode - Data which has maximum frequency.

Maximum - Largest of Data Points.

Minimum - Smallest of Data Points

Range 
- Differences between i i
Maximum & Minimum

2
Sx m
Standard Deviation -
n-1
 MEAN & STANDARD DEVIATION
SQC& OR UNIT, PUNE

20 20

15

10
10

0
0
1 2 3 4 5
 WHY STANDARD DEVIATION?
SQC& OR UNIT, PUNE

 Range tells us only about two points,

Max and Min

 Standard Deviation tells us about the

relative distance of all points from the
average or mean.

 Variation is important to know

 HOW TO CALCULATE STANDARD DEVIATION?
SQC& OR UNIT, PUNE

2
X X - Mean ( X - Mean)
1
2
3
4
5
6
7
8
9
10
Sum
m
Mean
s
s

How is it used in Normal distribution?

 CHARACTERIZE NORMAL DISTRIBUTION
SQC& OR UNIT, PUNE

1s 1s
68.27%
2s 2s
95.45%
3s 3s
99.73%
Do you mean to say, whether you like it or not, 99.73%
of the observations will lie within Average +/- 3 s.d.?
 PROCESS CAPABILITY
SQC& OR UNIT, PUNE

Natural or inherent
3s variation experienced
in a process.
6s
3s Valid only when the
process is in a state
of statistical control
 PROCESS CAPABILITY
SQC& OR UNIT, PUNE

Example of Chemical Company

Process Capability
(6 s)
s = 6.8
6 s = 6 x 6.8
= 40.8

Mean = 85.8%
65.4 100.0
Standard Deviation = 6.8%
 COMPARISON
SQC& OR UNIT, PUNE

VARIATION
LSL USL
ACTUAL
PERFORMANCE
VS
SPECIFICATIONS
Or
Customer
Requirements
 COMPARISON
SQC& OR UNIT, PUNE

Scenario: Specified
tolerance (T) exactly
equals to the process
capability, i.e. 6s

6s 4 ft. wide vehicle running

with a 4 ft. wide road

Tolerance (T) 2700 ppm

LSL USL
Are you comfortable with the status quo? You
want to reduce the variability from s to s1?
 REDUCE THE VARIATION
SQC& OR UNIT, PUNE

Now, the situation is bit

comfortable. 3 ft. wide
vehicle running on 4 ft.
wide road.
Have 1 sigma gap from
the boundaries of the
1 s1 1 s1 road
6 s1 T = 8 s1

Tolerance (T)
LSL USL
64 ppm
Are you comfortable with the status quo? You still want
to further reduce the variability from s1 to s2?
 REDUCE THE VARIATION
SQC& OR UNIT, PUNE

2 ft. wide vehicle running

on 4 ft. wide road
Have 2 sigma gap from
the boundaries of the
road.
T=10 s2
2 s2 2 s2

6 s2
Tolerance (T) 0.6 ppm
LSL USL

Looking for reducing variability further s2 to s3 ?

 SIX SIGMA WITHOUT SHIFT
SQC& OR UNIT, PUNE

Oh, Yes!
You have got it !!!

T=12 s3
3 s3 3 s3

6 s3 0.002 ppm or
2 ppb
Tolerance (T)
LSL USL
Wait a bit! Every time you are assuming that average has always been
maintain at target. Is it practical?
What happens if the average shifts to a distance of 1.5 s3 from the
target to the worse side?
 SIX SIGMA WITH SHIFT (SHORT TERM)
SQC& OR UNIT, PUNE

Let us assume, average can go as away as 1.5 times s.d. from the
target.

T=12 s3
(with 1.5 s3 shift)

3.4 ppm
4.5 s3 1.5 s3 1.5 s3

6 s3
Tolerance (T)
LSL USL
 SIX SIGMA LEVEL Vs. PPM
SQC& OR UNIT, PUNE
 SIX SIGMA LEVEL Vs. PPM
SQC& OR UNIT, PUNE
 SIX SIGMA LEVEL: DIFFERENT CASES
SQC& OR UNIT, PUNE

IRS - Tax Advice (phone-in)

100K
(66810 ppm)
Restaurant Bills
Doctor Prescription Writing
10K Payroll Processing
Order Write-up
Average Journal Vouchers
Wire Transfers
Company(6210 ppm)
1K
Purchased Material
Lot Reject Rate Air Line Baggage
Handling
(233 ppm)
100

10

Best in Class
Domestic Airline
(3.4 ppm)
Flight Fatality Rate
1
2 3 4 5 6 (0.43 ppm)
7
SIGMA (with ±1.5 Sigma Shift)
 IMPACT OF SIX SIGMA LEVEL
SQC& OR UNIT, PUNE

ERROR 3 Sigma Process 6 Sigma Process

Wrong drug prescription 5400 every year. 1 in every 25 years
Unsafe drinking water 2 hours each month 1 second every 16 years.

No TV transmission 27 minutes each 6 seconds in year

week
Short or long landing at an 5 per day 1 in 10 years
average airport
Incorrect surgical 1350 per week 1 in 20 years
operations.
Articles of mail lost 54000 per hour 35 per year
SQC& OR UNIT, PUNE
Sweet fruit Process
Design limitations
4s

Bulk of fruits
process
characterization 3s

Low hanging
fruits
7 basic tools
2s
Logic and intuition
Quick hits

Increase of sigma requires an exponential decrease in rejections

 WHY SAMPLING?
SQC& OR UNIT, PUNE

Sampling is
Collecting a portion (sample) representatively of all
the data (population) and draw conclusion on the
population.

Why sample? Because looking at all the data may be

Too expensive.
Too time-consuming.
Destructive (e.g., taste tests).

Sound conclusions can often be drawn from a

relatively small amount of data.
 SAMPLE SIZE
SQC& OR UNIT, PUNE

The first question people often ask is “How many samples

do I need?” The answer is determined by four factors:
1. Type of data
• Discrete vs. continuous
2. What you want to do
• Describe a characteristic for a whole group (mean
or proportion)
– Within a certain precision (± ___ units)
3. What you guess the standard deviation
(or proportion) will be
4. How confident you want to be (usually 95%)
 SAMPLE SIZE
SQC& OR UNIT, PUNE

Purpose of Sample Formula*/ Minitab Commands

2
Estimate average
 2s 
(e.g., determine baseline n  
cycle time)
 
(Where  = precision: ± __ units)
Estimate proportion 2
 2
(e.g., determine baseline % n    p1 p
defective)  d
(Where d = precision: ± __ units)
 PRECISION [d]
SQC& OR UNIT, PUNE

 Precision is how narrow you want the range to be for an

estimate of a characteristic.
 Estimate cycle time within 2 days.
 Estimate percent defective within 3%.
 Use the symbol d to represent precision.
Traditional statistics refers to it as “delta,” hence the d.
 Precision is equal to half the width of a confidence
interval.
 A 95% CI = (48, 52) for cycle time (in days) means we
are 95% confident the interval from 48 days to 52 days
contains the average cycle time.
 Width of the CI = 4 days.
 Precision = d = 2 days (= estimate is within ± 2 days).
 PRECISION [d]
SQC& OR UNIT, PUNE

 To improve precision, increase sample size

(which incurs more cost).
 There is no clear-cut answer about how
much precision you need; the answer
depends on the business impact of using
the estimate.
 SAMPLE SIZE - RULE OF THUMB
SQC& OR UNIT, PUNE

Statistic or Recommended Minimum

Chart Sample Size (n)
2
 2s 
Average n 
 d 
2
 2
Proportion n    p1 p
 d

Frequency plot 50
(histogram)

Pareto chart 50

Scatter plot 24

Control chart 24
SQC& OR UNIT, PUNE

CHAPTER - 3.4
PROCESS CAPABILITY
 TRADITIONAL METHOD
SQC& OR UNIT, PUNE

Major Drawbacks

Time consuming and Detailed

Difficult to interpret
Fails to go in to specifics
Not amenable for comparisons

Not suitable for rigorous improvement

studies
 STATISTICAL METHOD
SQC& OR UNIT, PUNE

Statistical way of understanding a

process

No two things will be alike

Variation
(deviation from target, set value or expectation)
 STATISTICAL METHOD
SQC& OR UNIT, PUNE

Understanding & Measuring

Variation

6σ
Total Process Variation = 6σ
where σ : Standard Deviation
SQC& OR UNIT, PUNE
How to use PC

Variation Tolerance
SQC& OR UNIT, PUNE
How to use PC

6σ < T

Variation (6σ)

Tolerance (T)
SQC& OR UNIT, PUNE
How to use PC

6σ = T

Variation (6σ) Tolerance (T)

SQC& OR UNIT, PUNE
6σ > T
How to use PC

Variation (6σ)
Tolerance (T)
 PROCESS CAPABILITY INDICES
SQC& OR UNIT, PUNE

For Stable Process:

Potential Capability (Cp) Achieved (Cp)

Cp  (USL - LSL) Cpu  (USL - X)

6s
or T 3s
6s Cpl  (X - LSL)
3s
C  Min Cpu, Cpl
 
 
 
pk 



 PROCESS CAPABILITY INDICES
SQC& OR UNIT, PUNE

For Unstable Process:

Potential Capability (Pp) Achieved (Pp)

Pp  (USL - LSL) Ppu  (USL - X)

6s
or T 3s
6s Ppl  (X - LSL)
3s
P  Min Ppu, Ppl
 
 
 
pk 



 PC Vs T
SQC& OR UNIT, PUNE

Process Capability
=
Tolerance
or
PC = T

6σ or
Cp = 1
Tolerance (T) Cpk = 1
LSL USL
 PC Vs T
SQC& OR UNIT, PUNE

Process Capability
<
Tolerance

1σ 1σ
6σ < T or T = 8σ
6σ Cp = 1.33
Cpk = 1.33
Tolerance (T)
LSL USL
 PC Vs T
SQC& OR UNIT, PUNE

Process Capability
<
Tolerance
6σ < T or T = 8σ
Cp = 1.33
6σ Process off centered
Cpk < 1.33
Tolerance (T)
LSL USL
 PC Vs T
SQC& OR UNIT, PUNE

Total Process
Variation
<
Tolerance
2σ 2σ
6σ < T or T = 10σ
6σ
Cp = 1.67

Tolerance (T) Cpk = 1.67

LSL USL
 PC Vs T
SQC& OR UNIT, PUNE

Total Process
Variation
<
Tolerance

6σ < T or T = 10σ
6σ Cp = 1.67
Process off centered
Tolerance (T)
LSL USL
Cpk < 1
 PC Vs T
SQC& OR UNIT, PUNE

Total Process
Variation
<
Tolerance
3σ 3σ
6σ < T or T = 12σ
6σ
Cp = 2.00
Tolerance (T) Cpk = 2.00
LSL USL
 PC Vs T
SQC& OR UNIT, PUNE

Total Process
Variation
<
Tolerance
6σ < T or T = 12σ

6σ Cp = 2.00
Process off centered
Tolerance (T)
Cpk < 1.00
LSL USL
 CTQ - SELECTIVITY
SQC& OR UNIT, PUNE

CTQ :Selectiv ity Mean: 0.434 Sigma:0.047

Specif ications: LSL=0.35 Nominal=0.50USL=0.65
Process Capability Indices Cp=1.07 Cpk=0.64

LSL NOMINAL USL

40

35

30

25
Frequency

20

15

10

0
0.25 0.30 0.35 0.40 0.45 0.50 0.55 0.60 0.65
 PROCESS CAPABILITY - SHORT & LONG TERM
SQC& OR UNIT, PUNE

Short term Capability

Estimated over a short period without
much deviations or departures in the
process.
Long term Capability
Estimated over a period during which
process deviations or departures are
also present.
 HOW TO CALCULATE PROCESS SIGMA
SQC& OR UNIT, PUNE

1) Compute actual First Pass Yield

Yield = (
1–
defects
total opportunities ) x 100
(total opportunities = number of units x opportunities per unit)

2) Look up Yield in Process Sigma Table

100 units
EXAMPLE: 5 opportunities
7 defects

7
Yield = 1 –( 5 x 100 ) x 100

= (1 – .014) x 100 = 98.6% = 3.7 s

 HOW TO CALCULATE PROCESS SIGMA
SQC& OR UNIT, PUNE

1. Determine number of defect opportunities O=

per unit

2. Determine number of units processed N=

3. Determine total number of defects D=

made (include defects made and later fixed)

D
4. Calculate Defects Per Opportunity DPO = =
NxO

5. Calculate Yield Yield = (1-DPO) x 100 =

6. Look up Sigma in the Process Sigma Table

Process Sigma=
SQC& OR UNIT, PUNE
 HOW TO CALCULATE PROCESS SIGMA
SQC& OR UNIT, PUNE

1. Determine number of defect opportunities O=

5
______
per unit

2. Determine number of units processed N= 100

______

3. Determine total number of defects D= 7

______
made (include defects made and later fixed)
D .014
4. Calculate Defects Per Opportunity DPO= = ______
NxO

5. Calculate Yield Yield = (1-DPO) x 100 = 98.6

______

6. Look up Process sigma in the Process Sigma Table

Process Sigma =
3.7
______
 HOW TO CALCULATE PROCESS SIGMA
SQC& OR UNIT, PUNE

Enter process sigma level, compute PPM

Process Sigma Level -> 1.9
PPM 344578.3
Percent 34.45783% 1000000*(1-NORMSDIST(B2-1.5))

Enter percent, compute PPM and process sigma level

Percent -> 1.00%
PPM 10000
Process Sigma Level 3.8 -(NORMINV((B8/1000000),0,1)-1.5)

Enter DPMO, compute process sigma level

Process Sigma Level -> 3.192577
DPMO 45268.0 -NORMSINV($B$14/1000000)+1.5
 HOW TO CALCULATE PROCESS SIGMA
SQC& OR UNIT, PUNE

A process that is “six sigma capable” will have the

specifications ±6 short-term standard deviations away from
the centered process
±1.5s 4.5s

3.4ppm
3.4ppm
short-term
Short-term
distribution
distribution

-6s -3s 0 3s 6s

± Six Sigma Design

The “6” in the short-term “6 sigma” has the 1.5 shift built in;
subtract the 1.5 to get long-term capability:
6 – 1.5 = 4.5
 NORMAL DATA, SINGLE SIDED
SQC& OR UNIT, PUNE

USL of 24 (LSL=0, Boundary)

Process Capability Analysis for Viscosity

LB USL
Process Data
USL 24.0000
Within
Target *
Lower Bound 0.0000 Overall
Mean 23.9624
Sample N 50
StDev (Within) 1.91051
StDev (Overall) 2.00174

Potential (Within) Capability

Cp *
CPU 0.01
CPL *
Cpk 0.01
0 5 10 15 20 25 30
Cpm *

Overall Capability Observed Performance Exp. "Within" Performance Exp. "Overall" Performance
Pp * PPM < LB 0.00 PPM < LB * PPM < LB *
PPU 0.01 PPM > USL 500000.00 PPM > USL 492140.99 PPM > USL 492499.13
PPL * PPM Total 500000.00 PPM Total 492140.99 PPM Total 492499.13
Ppk 0.01

How do we feel about this process’ capability?

 WHAT IS THE DIFFERENCE?
SQC& OR UNIT, PUNE

• ‘LB’ replaces ‘LSL’ in the information printed

below the histogram

• PPM < LB
– EXPECTED PPM < LB is impossible (Minitab
prints the missing value, *)
– OBSERVED PPM < LB should be zero!

• Check for a data entry error if OBSERVED

PPM < LB is not zero!
 NORMAL DATA, SINGLE SIDED
SQC& OR UNIT, PUNE

• Minitab assumes that the Target is midway

between the Upper and Lower Spec limits
• What if we are targeting a value other than the
midpoint?
• The measure we will get for process capability
is the Cpm
• We want the Cpm to approach 4.00
– This is a measure of centering on the TARGET,
not the middle of the spec
• Let’s take a look….
 NON-CENTRAL TARGET VALUE
SQC& OR UNIT, PUNE
USL  LSL
Target 
2
• The calculation of Cpk assumes that the
TARGET is halfway between LSL and USL:
USL  LSL
C pm 
 X T 
2
i
6
n 1
• If the Target is not in the middle, then Cpk is
not appropriate!

• A better capability statistic in this situation is

Cpm:
 NON-CENTRAL TARGET VALUE
SQC& OR UNIT, PUNE

• Another (mathematically equivalent) way to

calculate Cpm is as follows:

USL  LSL
C pm 
6 s 2  m  T 
2

• This formula reduces to the ordinary definition

of Cp, if Bias m  T  = 0.

• The quantity in the denominator expresses the

cost of variation (“Taguchi Loss Function”)
 ADD CPM TO OUR CAPABILITY OUTPUT
SQC& OR UNIT, PUNE

STEP 1. Input the

specs:
LSL = 0
USL = 24

STEP 2. Click the

“Options” button

Note:
If a spec is a
‘Boundary,’
Minitab will not
calculate Cpm!
 ADD CPM TO OUR CAPABILITY OUTPUT
SQC& OR UNIT, PUNE

STEP 3.
Input the
target.

Target = 8.

STEP 4. Click ‘OK’ twice to generate the

capability analysis.
 CAPABILITY OUTPUT WITH Cpm
SQC& OR UNIT, PUNE

LSL = 0, USL = 12, TARGET = 4

Process Capability Analysis for Viscosity

LSL Target USL

Process Data
USL 24.0000
Within
Target 8.0000
LSL 0.0000 Overall
Mean 23.9624
Sample N 50
StDev (Within) 1.91051
StDev (Overall) 2.00174

Potential (Within) Capability

Cp 2.09
CPU 0.01
CPL 4.18
Cpk 0.01
0 5 10 15 20 25 30
Cpm 0.25

Overall Capability Observed Performance Exp. "Within" Performance Exp. "Overall" Performance
Pp 2.00 PPM < LSL 0.00 PPM < LSL 0.00 PPM < LSL 0.00
PPU 0.01 PPM > USL 500000.00 PPM > USL 492140.99 PPM > USL 492499.13
PPL 3.99 PPM Total 500000.00 PPM Total 492140.99 PPM Total 492499.13
Ppk 0.01
 HOW SHOULD WE IMPROVE THE PROCESS?

SQC& OR UNIT, PUNE

• Process mean = ___________.

• Process target = ___________.
• Is the process on-target? ___________
• What can we do to improve the process?
___________________________________

• Capability analysis of the improved process

is shown on the next page.
 AN IMPROVED PROCESS (ON-TARGET!)
SQC& OR UNIT, PUNE

Process Capability Analysis for Viscosity Sh

LSL Target USL

Process Data
USL 24.0000
Within
Target 8.0000
LSL 0.0000 Overall
Mean 7.9624
Sample N 50
StDev (Within) 1.91051
StDev (Overall) 2.00174

Potential (Within) Capability

Cp 2.09
CPU 2.80
CPL 1.39
Cpk 1.39
0 5 10 15 20 25
Cpm 2.01

Overall Capability Observed Performance Exp. "Within" Performance Exp. "Overall" Performance
Pp 2.00 PPM < LSL 0.00 PPM < LSL 15.39 PPM < LSL 34.79
PPU 2.67 PPM > USL 0.00 PPM > USL 0.00 PPM > USL 0.00
PPL 1.33 PPM Total 0.00 PPM Total 15.39 PPM Total 34.79
Ppk 1.33
 WHAT ABOUT NON-NORMAL DATA?
SQC& OR UNIT, PUNE

• All of these discussions are great, but what about Non-normal

data?
• As we improve our process and Viscosity is getting closer to
our lower limit, we can expect to see non-Normal data
• We can now use a method called Transformation of the data
– Transformation is like using a different scale
– We must transform the data AND specification limits for
this to be valid
• NOTE: Only uni-modal data should be transformed!
• Why? _______________
 NON-NORMAL DATA
SQC& OR UNIT, PUNE

• Let’s look at Improved Viscosity data

• This type of distribution is typical of a
single sided spec item where Normal
there is aPlot
Probability

15
boundary .999
.99
.95

Probability
Frequency

10 .80
.50
.20
5 .05
.01
.001
0
0 5 10 15
0.0 1.5 3.0 4.5 6.0 7.5 9.0 10.5 12.0 13.5 15.0
Improved Vis
Improved Viscosity Average: 3.52129 Anderson-Darling Normality Test
StDev: 2.99849 A-Squared: 1.610
N: 50 P-Value: 0.000
 CAPABILITY STUDY
SQC& OR UNIT, PUNE

• If we perform a capability study with the non-

normal data, we can still get results …. but are
they correct?
Process Capability Analysis for Improved Vis

LB USL
Process Data
USL 24.0000
Within
Target *
Lower Bound 0.0000 Overall
Mean 3.5213
Sample N 50
StDev (Within) 2.60905
StDev (Overall) 3.01382

Potential (Within) Capability

Cp *
CPU 2.62
CPL *
Cpk 2.62
-5 0 5 10 15 20 25
Cpm *

Overall Capability Observed Performance Exp. "Within" Performance Exp. "Overall" Performance
Pp * PPM < LB 0.00 PPM < LB * PPM < LB *
PPU 2.26 PPM > USL 0.00 PPM > USL 0.00 PPM > USL 0.00
PPL * PPM Total 0.00 PPM Total 0.00 PPM Total 0.00
Ppk 2.26
 THE BOX-COX TRANSFORMATION
SQC& OR UNIT, PUNE
• We can use the Box-Cox Transformation to determine a transformation value
• When you use the Box-Cox Transformation, Minitab estimates a  (lambda)
value
– Minitab presents the  value which minimizes the s of a standardized transformed
variable
– The resulting transformation is Y when   0 and Log  when  = 0
• This method searches through many types of transformations
 THE BOX-COX TRANSFORMATION
SQC& OR UNIT, PUNE
 THE BOX-COX TRANSFORMATION

SQC& OR UNIT, PUNE

• The output is a graph which displays:

– the best estimate Plot
Box-Coxof forfor theVistransformation
Improved

– two closely competing values of  95% Confidence Interval

9 Last Iteration Info

– a 95% confidence interval for the estimated value

8
Lambda
Low 0.392
StDev
2.299

of  7
Est
Up
0.449
0.506
2.301
2.309

6
StDev

2
-1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0
Lambda
 SO WHAT?
SQC& OR UNIT, PUNE

• Now we can use the  value to transform the data

and conduct the Capability Study
• NOTE: We must transform the Target value first!

Constant Value Name

K1 8.00000 Target
K2 1.86348 Target Xfrm
 CAPABILITY STUDY
SQC& OR UNIT, PUNE
 CAPABILITY STUDY
SQC& OR UNIT, PUNE

Process Capability Analysis for Improved Vis

Box-Cox Transformation, With Lambda = 0.449
LB* Target* USL*
Process Data
USL 24.0000
USL* 4.1660 Within
Target 8.0000
Target* 2.5438 Overall
Lower Bound 0.0000
Lower Bound* 0.0000
Mean 3.5213
Mean* 1.6043
Sample N 50
StDev (Within) 2.60905
StDev* (Within) 0.65506
StDev (Overall) 3.01382
StDev* (Overall) 0.66598
Potential (Within) Capability
Cp *
CPU 1.30
CPL *
Cpk 1.30 0 1 2 3 4

Cpm *

Overall Capability Observed Performance Exp. "Within" Performance Exp. "Overall" Performance
Pp * PPM < LB 0.00 PPM < LB* * PPM < LB* *
PPU 1.28 PPM > USL 0.00 PPM > USL* 46.06 PPM > USL* 59.93
PPL * PPM Total 0.00 PPM Total 46.06 PPM Total 59.93
Ppk 1.28
 WHAT’S CHANGED?
SQC& OR UNIT, PUNE Process Capability Analysis for Im
•Any item with anBox-Cox
asterisk Transformation,
next to it is for With Lamb
the transformed data
LB* Target*
Process
Process Capability Analysis for Improved
• Notice that histogram for the original data
Data Vis
USL 24.0000
USL* Box-Cox is shown
4.1660 as a referenceWith Lambda = 0.449
Transformation,
Target 8.0000
LB*
Target* • All capability indices are now
2.5438 for the
Target*
Process Data Lower Bound Transformed Data ONLY!
0.0000
USL 24.0000 Lower Bound* 0.0000
USL* 4.1660 Mean • 3.5213
Let’s compare the original data with the
Target 8.0000 Mean* 1.6043
Sample N transformed
50 data...
Target* 2.5438
StDev (Within) 2.60905
Lower Bound 0.0000
StDev* (Within) 0.65506
Lower Bound* 0.0000 StDev (Overall) 3.01382
Mean 3.5213 StDev* (Overall) 0.66598
Mean* 1.6043
Sample N 50 Potential (Within) Capability
Cp *
StDev (Within) 2.60905 CPU 1.30
StDev* (Within) 0.65506 CPL *
StDev (Overall) 3.01382 Cpk 1.30 0 1 2
StDev* (Overall) 0.66598
Cpm *
Potential (Within) Capability
Cp * Overall Capability Observed Performance Exp. "Within" Performanc
CPU 1.30 Pp * PPM < LB 0.00 PPM < LB*
PPU 1.28 PPM > USL 0.00 PPM > USL* 46
CPL *
PPL 0 * 1 PPM Total 2 0.00 PPM
3 Total 46
Cpk 1.30 Ppk 1.28
Cpm *
 CAPABILITY SIX PACK
SQC& OR UNIT, PUNE

• By the way, we can also conduct a Sixpack with the

same methods…
• Please note the lack of the original data histogram but
the Title changes
Process Capability Sixpack for Improved Vis
Box-Cox Transformation With Lambda = 0.449
Individual and MR Chart Capability Histogram
UCL=3.570
Individual Value

2
Mean=1.604
1

0
LCL=-0.3608
0 2 4
Obser. 0 10 20 30 40 50
Normal Prob Plot
2.4 UCL=2.414
Mov.Range

1.6

0.8 R=0.7389

0.0 LCL=0
0.0 1.5 3.0

Last 25 Observations Within Capability Plot

StDev: 0.655062 Process Tolerance
3 Cp: 1.06 Within
Cpk: 0.82 I I
Values

2 Overall
Overall I I
1 StDev: 0.665976 Specifications
Pp: 1.04 I I I
0 Ppk: 0.80 0 T 24
30 40 50
Cpm: 0.60
Observation Number
 STUDY COMPARISONS
SQC& OR UNIT, PUNE

• Non-transformed data results:

– Cp = ___ Cpk = ___
– Pp = ___ Ppk = ___

• Transformed data results:

– Cp = ___ Cpk = ___
– Pp = ___ Ppk = ___
•So which best describes the process ?
•_______________
•What implications does this have when we
report capability data to our boss and our
customers?
 WARNING !!!
SQC& OR UNIT, PUNE

Statistical Assumptions Made In Capability Studies

1. Data Comes From A Stable Process

If Not, Work Towards Getting It In Control

Don’t despair, you can still make some assumptions

about your process in the interim

2. Has A Normal Distribution

If Not, Transform It (Only if mono-modal data)

If Items #1 and #2 aren’t met, results will be misleading

SQC& OR UNIT, PUNE
HINTS

Use hard limits only when they make practical sense.

Box-Cox transformations doesn’t work on Zero values.

Only transform when you have a statistical reason.

Transformation doesn’t only apply to capability, it may

apply in Multi-Vari or DOE when you are trying to get
your data to meet normality assumptions.

______________________________________
______________________________________
 BASELINE STATUS
SQC& OR UNIT, PUNE

• Baseline Status will appear in the Excel sheet as an

augmentation in the CTW tree. Baseline status will
include as under:
– Total Number of observations
– Trend analysis through time series plot
– Selection of data
– Histogram plot of select data
– Normality test of select data
– Adequacy of sample size of selected data
– Average, standard deviation, Cp, Cpk, Pp, Ppk, dpmo, Sigma
level- Short-term and Long-term
• The baseline status is augmented with CTQ tree as
provided in the next slide.
 BASELINE STATUS
SQC& OR UNIT, PUNE
Code Y
CTQs Cycle time
Type of Data Variable
Baseline Status:
No. of observations: 18
Time Series Plot Apparently chronic
Histogram: Apparently by-modal and
hence may not be normal
Normality test Not normal
Average 251.6
s.d. 96.2
Pp
Ppk -0.66
Select homogenious group:

No. of observations: 8
Normality test Normal
Average 151.4
s.d. 24.4
Adequacy of sample size with 17.3
precision level:
Cp
Cpk -1.25
dpmo 500000
Sigma level- short term: 1.5
Sigma level- long term: 0
Baseline: I
378
340
154
342
263
 EXERCISE-10: BASELINE STATUS
SQC& OR UNIT, PUNE

• Create the CTQ tree with QC Support System,

Measurement Support system, Baseline Status for your
project as per the following:
QC-Support Code Z1 Z2 Y Y1 X
System CTQs
Type of Data
UoM
USL
Target
LSL
Min freq. of data:

Record ref.
Responsibility
Measurement Accuracy:
Support System
Instrument:
Make
L.C.
Calibration
Bias
Precision:
%R&R
Sigma R&R
 EXERCISE-10: BASELINE STATUS
SQC& OR UNIT, PUNE

• Baseline Status: (contd…)

Baseline Status:
No. of
observations:
Time Series Plot
Histogram:
Normality test
Average
s.d.
Pp
Ppk
Select
homogenious
group:
No. of
observations:
Normality test
Average
s.d.
Adequacy of
sample size with
precision level:
Cp
Cpk
dpmo
Sigma level- short
term:
Sigma level- long
term:
 EXERCISE-10: BASELINE STATUS
SQC& OR UNIT, PUNE

• Baseline Status: (contd…)

Baseline Data: Unit