Six Sigma

D M A I C
Six Sigma Study Material
SAMIR SHAH Slide 1 Six Sigma Module1-

Old concept of quality
Past concepts of quality focused on “conformance to standards”. This definition

assumed that as long as the company produced quality products and services, their
performance standard was correct regardless of how those standards were met.
Moreover, setting of standards and measurement of performance was mainly
confined to the production areas and the commercial and other service functions
were managed through command and control.

New concept of quality
In Six Sigma, the concept of quality encompasses manufacturing, commercial and other
service functions of an organization because all these functions directly or indirectly affect
products/service quality & customer satisfaction. It also embodies a structured system of
capturing all types of errors and their quantified measurement for subsequent analysis and
improvement.
Under the new approach, quality is a state in which Value enrichment is realized for the
customer and provider in every aspect of business relationship.

Visible and hidden costs
•Scrap
•Rework
Visible costs
•Warranty costs
• Conversion efficiency of materials
• Inadequate resource utilization
• Excessive use of material

• Cost of redesign and re-inspection Hidden Costs
• Cost of resolving customer problems
• Lost customers / Goodwill
• High inventory

Global acceptability The world over, organizations seeking
better ways of improving their profitability
have adopted Six Sigma because of its
ability in giving quick tangible results.
These organizations include General
Electric, Allied signal, Honda, Sony,
Canon, Polaroid, Texas instruments and
many more fortune 500 companies across
the globe. Indian Companies
ISPAT
TATA Special Steel
Whirlpool
Motorola
GE
Honeywell
ICICI Bank
Standard Chartered
Satyam
Wipro
Pioneer of six sigma = Bob Galvin (CEO-Motorola) Samtel
LG Electronics
Six Sigma originated from Motorola in 1979
Samsung
TATA Johnson Controls
Cummins
Around the World
Six Sigma is the most important initiative GE has

ever taken…it is part of the genetic code of our future leadership.
-Jack Welch, Former CEO,
General Electric
You can see the “before” & “after” of an organization

when Six Sigma grabs hold & takes place.
-Richard Johnson, Director of Six Sigma,
Allied Signal
Successful leaders have made Six Sigma their way of conducting business.
-Azim Premji, Chairman,
Wipro Limited

Six Sigma Application
Why : Business Strategy, World Class Performance
Where : Customer Critical Criterion
What : Defect Reduction
How : Projects, DMAIC, People Development, Tools
Who : Black Belts, Green Belts, Team Members
When : Usually 3/4 months per project

Why Six Sigma….
Continuos defect reduction in Products & Services

Six Sigma approach integrates the human and process
elements of improvement to produce breakthrough
results
Enhanced Customer Focus
Process Sustenance
Project based improvement, with visible milestones
Personnel Development
Strong leadership
Training
Project teams
Culture change
Global acceptance to Quality System

Knowing characteristics critical to customer
Value enrichment for the company and the customer is done by measurement based
approach of Six Sigma, which enables us to find :-
Vast areas of business about which we do not know enough.
When we do not know about a parameter or a characteristic which are important to
customers, we do not value it.
If we do not value a parameter or a characteristic, we do not measure it.
If we do not measure, we can not improve.

What is Sigma ……
The term ‘Sigma’ taken from the Greek alphabet, is used

to designate the distribution or spread about the mean
(average) of any parameter of product, process or
procedure.
Sigma capability is a metric which indicates how the

process is behaving. Higher the Sigma level, better the
capability of the process to produce defect free work and
vice versa.

What is Six Sigma ……
Six Sigma is an approach to achieve customer satisfaction

& business excellence through a systematic problem
solving methodology aiming at identifying & reducing the
process variation to achieve optimized process output.
Organizations have no choice but to offer the best

quality products and services at the minimum cost -
which is the purpose and objective of Six Sigma.

What is Six Sigma?
Lower Upper
Specification Limit Mean / Target Specification Limit
High 66807 Defects

Probability Per Million
of Failure Opportunities
1
3 1’s 3 1’s

Higher
Higherthese
this
numbers,
number,
Lower the
chance of Much Lower 3.4 Defects
producing a Probability Per Million
defect of Failure Opportunities
1
6 2’s 6 2’s

Six Sigma is…..
3.4 defects per 10,00,000 opportunities
That means a process efficiency of 99.99966%

In other words, process standard deviation () should be so less that your process performance can fit
12 within the customer specified limits. So, no matter how much your process deviates from target, you
always meet what customer wants.
Six Sigma is a high performance data driven approach focused on analyzing
On Center the root causes of business problems and solving them. It ties the outputs of a
Small Spread business directly to customer.
Sigma is a Statistical unit of measurement which reflects the capability of the
Six Sigma is about
process.
Meeting the Target,
Consistently Sigma reflects how well the process is performing. The higher the value the
less chance of producing a defect.
A defect is anything that causes Customer Dissatisfaction.
Cp ; Cpk ; DPU ; DPMO are some of the measures to calculate Sigma.
LSL T USL
Eliminate variation , defects & its opportunity .
Tool for : > Customer focus
Breakthrough improvement
Continuous improvement
Practical Problem Statistical Problem
Robust Process
People Involvement
y  f ( x1 , x 2 , ... , x k )
Practical Solution Statistical Solution

Six Sigma Organizational structure…
Steering Committee
Six Sigma Black Belt / Core Team

Champion
Sponsor
Black Belt
Green Belt
Team Members

Teaming Roles
Champion ( GM )
Is the Head of the Business Unit / Division
Approves Green Belt project along with the Sponsor
Reviews projects periodically with Sponsor
Adds value in project reviews & Clears road blocks for the teams.
Has the overall responsibility for the project closure & recognition of successful Project
Team

Teaming Roles
Sponsor ( Manager / Section Head )
Selects Green Belt project along with the Black Belt
Must identify the project deliverable for Green Belt in Project Charter
Reviews projects of Green Belt frequently with Black Belt
Provide resources to Green Belt Leader for accomplishment of Target.
Monitor progress of Green belt project & generate MIS for Top Management.

Teaming Roles
Black Belt ( GIHL Six Sigma Experts )
Trains others in Six Sigma methodologies & concepts
Conducts programs on Statistical tools
Interact along with the Business / Functional Unit Head and helps project selection
Provides Application Assistance by facilitating team discussions
Helps review projects with Business Unit Head
Facilitation of Green Belt Projects.

Teaming Roles
Green Belt ( Engineers / Supervisors )
Is the Team Leader for a Green Belt Project
Selects other members of his project team
Uses Six sigma tools for achieving the goal of project
Defines the roles and responsibilities for each team member
Identifies training requirements for team along with Black Belt
Work out the Financial Score Card along with his Sponsor / Chief Financial Officer

Teaming Roles
Team Members ( Technicians / Operators )
A Team Member is chosen for a special skill or competence
Team Members help in analysis of the project goal.
Team Members drive the project to completion
Guidelines for team members
Up to 5 people
People who have stake in the process
People who are benefited by removal of pain area
People who have complementary skills
People from same location
Guest members

Six Sigma
methodologies

Project Methodologies
Six Sigma
Defect Reduction Cycle Time Reduction
DCAM DMAIC DSSS TQSS CFPM

Design for Define
Customer Measure Developing Transactional
Satisfaction Cross Function
Analyze Six Sigma Quality using
And Process Mapping
Improve Software Six Sigma
Manufactur-
-ability Control
Design & Transaction Software Transaction

Cross-function
Manufacture Cross-function Development Uni-function
Note: DMAIC- is most popular methodology for continuous process industry

DMAIC Methodology of Six Sigma
DEFINE
What do we work on?
How does it tie into business goals?
What are the internal deliverables?
DMAIC is one of the Six Sigma
methodologies
MEASURE
What are the specification limits?
What is a defect? It’s best applicable to continuous
improvement in cross-functional
environment
ANALYZE
What is the process capability?
How much you want to improve it? DMAIC applies to existing processes
that need improvement
IMPROVE
What are the factors that govern the CTQs?
Where should they be set?
CONTROL
How can results be sustained?

Sigma Level Vs DPMO
Sigma Level DPMO Yield

6 Sigma 3.4 99.99966%
5 Sigma 233 99.97673%
4 Sigma 6,210 99.37903%
3 Sigma 66,807 93.31928%
2 Sigma 3,08,537 69.14625%
1.5 Sigma 5,00,000 50.00000%
Sigma() is a measurement of quality to know how effectively we

eliminate defects & variations in a process.
Sigma Level: A measure of process performance.(calculated from DPMO).
DPMO : Defects per million opportunity.
Yield : Defect free percentage.
Why 99% is Not Good Enough?
@ 99% @ Six Sigma

More than 110,000 newborn babies Less than 38 newborn babies accidentally
accidentally dropped by doctors and nurses dropped by doctors and nurses each year
each year
No electricity for 85 hours each year No electricity for 9 minutes in 5 years
No television transmission for nearly 64 No television transmission for 11 minutes in

minutes per week 10 years
Four short or long landings per day One short or long landing every two years
16 railway accidents per day 2 railway accidents per year
16 minutes per week of unsafe water supply 1.4 minutes of unsafe water every 5 years

Cost of Poor Quality
30%__
There is enormous
25%__ opportunity to Improve…..
20%__
Cost of Poor Quality 15%__

as a % of Sales
10%__
Good
Companies
5%__
0% __
World class
Companies
| | | | |
3.4 233 6,210 66,807 308,537 DPMO
6 5 4 3 2 Sigma Multiple

Key Focus Areas for DMAIC Applications
Manufacturing
Commercial
Sales & Marketing
Sourcing ( Vendor development )
Information Systems
Human Resources
Any other?

Six Sigma Benefits
Continuous Defect Reduction in Products & Services
Enhanced Customer Focus
Process Sustenance
Performance Dashboards & Metrics
Project Based Improvement, with Visible Milestones
Sustainable Competitive Edge
Global Acceptance of the Quality System

Control
Improve
Analyze
Measure
DEFINE
DMAIC Steps
Step 0 Select a Project

D
Step 1 Establish Performance Parameters
Step 2 Validate Measurement System for ‘Y’ M
Step 3 Establish Process Baseline
Step 4
Step 5
Define Performance Goals
Identify Variation Sources
A
Step 6 Explore Potential Causes
Step 7
Step 8
Establish Variable Relationship
Design Operating Limits
I
Step 9 Validate Measurement System for ‘X’
Step 10
Step 11
Verify Process Improvement
Implement Process Controls
C
Introduction - DEFINE
Success of any Six Sigma endeavor relies heavily on project definition & selection
This is true for any project – no matter what methodology you select
A project must address a stated customer requirement
Your competitive advantage is to respond to the customer requirements, faster than

your competitor, with highest quality, at lowest possible cost
Everyone in the organization has a role in driving Six Sigma

Deliverables of Step 0
0.1 Define Business CTQ Sponsor,Champion
0.2 Define your Customer Sponsor, Champion,BB
0.3 Explore Customer CTQ Sponsor, Champion,BB
0.4 Define Internal CTQ / CTP’s Sponsor, Champion,BB
0.5 Select a Project & Green Belt Sponsor,Champion,BB
0.6 Scope the Project Sponsor, Team leader,BB
0.7 Develop Team Charter Sponsor,Team leader,BB

0.8 Quantify Benefits Sp.,Team leader,BB,Finance
0.9 Sign-off with Sponsor / Champion Sponsor, Champion
0.10 Kick-off the Project Sponsor ,Champion,

CTQ (Critical to Quality)
A CTQ is a
Product or Service Business CTQ

characteristic
Customer CTQ
that satisfies a
Customer Requirement
Internal CTQ
OR
Process Requirement

Who is your customer?
A customer is someone who

Uses your product or service
Decides to buy your product or service
Pays for your product or service
Gets impacted by your product or service
Internal & External customers
Primary & Secondary customers
Key concepts
Different markets / segments

Different customer requirements
Product / Service chain

Customer CTQs
Six Sigma begins with the customer
Customers find it easier to define what they do not want
Customer CTQs are defined by customers
Sources of customer CTQs
Survey results
Service reviews
Meetings

Examples of Customer CTQs
A Car Purchaser A TV Purchaser

• Good Acceleration • Flat Screen
• Spacious • Good Sound
• Affordable • NTSC / PAL Compatible
• Power Steering • Affordable
• Loan Facility • Good After Sales Service
A Caller to Help Desk A Prospective Employee

• Quick Answering • Good Salary
• Courteous Response • Location Preference
• Quick Problem Resolution • Flexible Working Hours
• ESOPs
• Separate Cabin

Steps in Project Selection
Identify pain areas / project themes in internal CTQs / CBPs – Brainstorming
Prioritize pain areas / project themes & – Multi-voting / Theme selection matrix
Develop fact-based business case – Why do this project?
Select a Green Belt
9 Key Questions
1. How important is it to your customer?

2. Is there a champion who feels that the project is important?
3. Is it a part of GB’s current job responsibilities & objectives?
4. Is the CTQ measurable?
5. Is data available or easily generated?
6. Are the benefits easily measurable?
7. Is the process stabilized or at least controllable?
8. Is the scope narrow enough to finish in 3 to 4 months?
9. Does the project have a high potential for replication?

What is Scoping?
It’s an attempt to define what will be covered in the project deliverables
Scoping sharpens the focus of the project team
& sets the expectations right
Types of Scoping
Longitudinal Scoping
Lateral Scoping

Longitudinal Scoping
Longitudinal scoping is done on the length of the process

e.g. – From the receipt of PO till the delivery at the distributor's go-down
e.g. – From the time of customer reporting the complaint till final satisfaction confirmation
Mostly the ‘start’ & ‘end’ points are baton change points
Points to remember:
There should not be too many baton changes with in the scoped process
The pain areas (identified at the time of project selection) must be within the selected scope
Between the “Start” and “End” of the process, there should be logical flow of units

Lateral Scoping
Lateral scoping is done on the breadth of the process

e.g. – All despatches from North & South regions
e.g. – Calls received during general shift
One or more of the following are covered here:
What all kinds of units, the process will cover
In what situations the process is valid
What are the qualifiers for the transactions
What domains does the process cover
In what geographical / functional areas , the process is valid

Scoping Techniques
Write inside the box what you think project covers

Write outside the box what you think project excludes Tool
Example for an ‘On time Despatch’ project
Inside-Outside
Longitudinal
Any IT related work
Starts after receipt of PO from customer
Any transit delays
Ends at the despatch from factory
Lateral Product damages
Despatches from Mumbai plant Billing errors

All despatches during shift A Non-availability of customer
Despatch through transporter X
Project boundary

Scope Verification
After the scoping has been done, following should be checked to validate the scope
If the scoped process is with in the control of the GB
If the pain area lies with in the scoped process
If the scoped process is narrow enough
If there are enough transactions to measure (at least 20 transactions per month are
recommended for effective measurements)
If the scope is still aligned with the Internal CTQ / CBP
If the scoped process would still result in achieving the objectives set by the Sponsor /
Champion.
Re-scoping may be needed later after spending considerable time on the project

Develop Team Charter
Fill this table with your Sponsor for the roles of each team member
Tool
Name of the team Functional

Sl. No. Responsibilities
member Role
Team Charter
2

Why Teaming?
There is a difference between ‘process owner’ & ‘process operator’
A team, as compared to an individual, is better equipped
to understand the holistic view of a process
to collect & possess all necessary data
with hands-on experience of process
to validate improvement plans
to implement all actions

Projection of Benefits
Six Sigma has a strong focus on money
Management buy-in is easier for tangible benefit projects
Tangible benefits could be of various types
Cost Reduction / Saving
Increase in Sales / Revenue
Enhanced productivity
Enhanced measurable Customer Satisfaction
Enhanced measurable Employee Satisfaction

Project Agreement Form
Champion’s / Sponsor sign-off ensures a common understanding of deliverables

It reflects his approval on team members & financials
Tool
A typical agreement form is as below
Project Agreement
Call a Team Meeting
Call a project team meeting to formally begin the project
Ask Champion to define his expectations from the project
Communicate objective of the project
Arrive at project time frame and deadlines
Present the teaming module, complete recommended exercises & develop role
clarity for each project team member
Get agreement from all members on their participation

Motivating the Team – Need Sharing
Why Bother?
Ensures a common understanding
Validates why the project is important and critical to do
Builds the momentum needed for the launch
What are we after?
A shared recognition, by both the team and the key constituents, of the need and logic of
the change sought
Dissatisfaction with status quo
Ability to frame the need as both an opportunity and threat

Create This Matrix with Your Team
Tool
Loss Opportunity
What are the What are the
Loss-Opportunity Matrix
short-term threats Immediate gains
Short Term if we don’t do if we are successful
the project? In the project?
What are the

What are the
Long Term long-term threats
long-term gains?
you can foresee?

Tollgate - Define
Signed-off Project Agreement Form
Loss-Opportunity Matrix
Elevator Speech

Control
Improve
Analyze
MEASURE
Define

D MAIC
Step 1
Establish
Performance
Parameters

DMAIC Steps

D
 Step 1 Establish Performance Parameters
Step 4
Step 5
A
Step 7
Step 8
I
Step 10
Step 11
C
Introduction - MEASURE
A robust measurement system forms the basis of any Six Sigma project
A measurement system has two characteristics 

Step 1 of DMAIC
Design of the measurement system
Accuracy of the measurement system

Step 2 of DMAIC

1.1 Map the As-is Process

What is a Process?
A process is a sequence of activities performed on an input to produce an output
A process must add value to the input
Every process has at least one supplier & one customer
Process Mapping is as critical to a manufacturing operation as it is to a transaction

SIPOC
Tool
P
Process
S I O C
SIPOC
1 2 5
Suppliers Inputs Outputs Customers

4
Process Boundary

Process Elements
Suppliers
Manage the supplier by giving clear specifications on requirements or data
Inputs
Data / unit required to execute the process
Process Boundary
Identified by the hand-off at the input (the start point of process) and the output (the end
point of the process)
Outputs
Output of a process creating a product or service that meets a customer need
In your project, output characterizes the pain area / project theme
For example, if pain area is truck loading during despatch, output measured for your
project could be time taken in loading
Customers
Users of the output

Versions of a Process
At least Three Versions
(Usually)
What You Think It Is... What It Actually Is... What You Would Like
It To Be...

Example of a SMS / Caster
S I Steel
O C
Making
BF/SIP HM/DRI
Process LM Caster
S I O C
Casting Process
SMS
LM Slab Mill

Process Mapping Nomenclature
Process Decision Data
Pre-defined Document Terminator

Process
Manual Manual
Delay Input
Operation

As-is Process Mapping
Gate In Security Gate In Document
IN entry put stamp entry Checking at
Taking gross
On paper weight
In SAP In Register W.B.
Go to O2 Instructions HP As per supplier If HP

Plant & intimate from shift crew Tanker Y
Send to their tank has
to shift crew for unloading type
storage space
LP N
If LP
N Y
Wait tank has Unload.
space
OUT Gate out Gate out Go to W.B. Go to O2 At W.B.

entry entry For documents Plant for for
In SAP In Register submission Wt. confirm Tare wt.

Total Cycle Time
Value added
Elapsed time ( no activity )
activity
Non-Value added
activity Time dimension of process
•Examine largest time delays for cause, e.g..

•Constrained resources •Non-value added activities
–Idle time
•Sequencing
–Queue time
•Lot Size –Setup
–Storage
•Defects –Transit
–Quality defects (yield)
•Develop plans to eliminate delays and reduce the cycle.
Benefits of Process Mapping
Tremendous value in having teams just discuss the process
Simple & visible structure for thinking through a complex process
Enables seeing the entire process as a team
Enables seeing that changes are not made in a vacuum and will carry through,
affecting the entire process down the line
Creates a framework for designing performance standards for your project
Magnifies non value-added areas or steps
Identifies cycle times of each step in the process
Helps re-examine (if needed) the scope and charter of your project

Key Concepts
Recall that customers are better off telling you what they do not want
A defect is an imperfection or deficiency in the output unit with respect to specifications

defined by the customer
Quality is absence of defects in the unit identified
Quality goes up as defects come down
Defects are defined by customers
A unit may have multiple defects depending upon customer CTQs
Defects & Defectives

What is a Specification?
A specification is a customer-defined tolerance for the unit characteristics
There may be two-sided specifications
Specifications form the basis of any defect measurement exercise
USL: Upper Specification Limit for a Performance

Parameter, anything above this is a defect.
LSL: Lower Specification Limit for a Performance

Parameter, anything below this is a defect.
Target: Ideally the middle point of USL & LSL.
LSL Target USL

Examples
Steel - HR Coil example
Unit: each HR coil received by the customer
Defect definition: a Coilt with out of thickness spec

Defective Coil
a Coil with less Hardness
a Coil with less weight / size
Pizza delivery example
Unit: each order placed
Defect definition: an order delivered after 20 minutes

an order not delivered hot
an order not delivered with salt & pepper
an order not delivered at all

1.4 Understand Data Characteristics

Types of Data
Discrete data
Data that can take a limited number of values
Examples
Number of orders delivered late
Number of days to resolve a problem
Number of ‘yes’ responses to a satisfaction survey
Number of matches won by A football team
Continuous Data
Data that be expressed in either fractions or whole numbers
Examples
Number of possible data points between 5 & 100
Temperature of the room
Exchange rate of a currency
Yield of a process

Continuous Data Characteristics
Location / Central Tendency
It is a measure of the center point of any data set
Spread / Dispersion
It is a measure of the spread of any data set around its center
Shape
It is a measure of symmetry of any data set around its center

Measuring the Location
Mean
Mean is the arithmetic average of all data points in a data set
Y1 + Y2 + Y3 + ………. + Yn
Y= Where n = number of data points
n
Mode
Mode is the most frequently occurring data point in a data set
Median
Median is the middle data point of a data set arranged in an ascending / descending order
Odd number of data points Even number of data points
Average

Measuring the Spread
Range
Range is the difference between the maximum & minimum data point
Variance / Standard Deviation
Variance & standard deviation measure how individual data points are spread around mean
( Y1 - Y )2 + ( Y2 – Y )2 + ……. + ( Yn – Y )2
Variance = s2 =
(n–1)
Standard Deviation = s = s2

Importance of Spread
C
B
A
Mean of Curve ‘A’ is more representative of its data set as compared to Curves ‘B’ & ‘C’
Spread outside the specifications may result in defects; this information is not
provided by mean
From a process perspective, individual customers are subject to different behaviors

of the process

Measuring the Shape
Symmetric Data set
It’s a data set in which spread of the data set around its mean is identical
For such a data set - mean = mode = median
Asymmetric Data set

Positive / Right skewed - high spread on the right side of the mean
Negative / Left skewed - high spread on the left side of the mean
Mode Mean Mean Mode
Median Median

The Complete Picture
On Center Off Center
Large Spread Large Spread
LSL T USL LSL T USL
Off Center On Center

Small Spread Small Spread
LSL T USL LSL T USL

Which Type of Data is Preferable?
Discrete data can detect changes up to about 4 – 4.5 σ multiple
At 4 σ multiple, large sample sizes are required to detect changes (>1000 observations)
Since continuous data measurements can be broken down, relatively smaller

samples are required for higher Sigma calculations
Average companies, which operate about 3 – 4 σ multiple, 80% of the measurements

are discrete
In a Six Sigma company, 80% of the measurements are continuous

1.5 Find Opportunities for Error

Opportunities for Error (OFE)
Opportunities for error in a process is the number of steps / tasks / actions in the
process, where there is a possibility of committing an error, that may result in a defect
OFE enables to compare the output quality of dissimilar processes
Concept of OFE is applicable only when defect measurement is discrete
This is because data, on whether or not a defect is created, is discrete type (yes / no)

1.6 Design Sampling Plan

Introduction to Sampling MAIC
D
We do sampling all the time
Populations & Samples Tool

Practical aspects – Cost & Time
Sampling is done to study a representative portion of population
Any term describing the characteristics of a sample is called statistic
Any term describing the characteristics of a population is called parameter
Sampling
Sample
Population

MAIC
D
Populations & Samples
Table 1.60 Population Sample
Collection of items Portion of the

Definition
being considered population chosen for study
‘Parameter’ ‘Statistic’
Population Size = N Sample Size = n

Characteristics
Population Mean = µ Sample Mean = X
Population Standard Deviation = σ Sample Standard Deviation = s

D MAIC
Statistic & Parameter
Sample Mean ‘ X ’ Population Mean ‘ µ ’
X1 + X 2 + X 3 + … + X n X1 + X 2 + X 3 + … + X N
n N
Sample Variance ‘ s2 ’ Population Variance ‘ σ2 ’
( X 1 - X )2 + ( X 2 – X ) 2 + … + ( X n – X ) 2 ( X1 - µ )2 + ( X2 – µ )2 + … + ( XN – µ )2
(n–1) (N)
Sample Standard Deviation ‘ s ’ Population Standard Deviation ‘ σ ’
s2 σ2
n = number of data points in the sample N = number of data points in the population

Types of Sampling D MAIC
Random Sampling or Probability Sampling
 All items in the population have an equal chance of being chosen in the sample
 Example: A customer satisfaction survey team picking the customers to be contacted at random
Stratified Random Sampling

 Randomly sample a proportionate number from each group.
 Example: picking male, female from a population.
Systematic sampling
 Sample every nth one
 Example : pick every 3rd or 5th item
 Subgroup Sampling
 Sample n units every t th time
 Example : 3 units every hour

Avoid Sampling Bias D MAIC
Bias occurs when systematic differences are introduced into the sample as a result of the sample
selection process
A biased sample would not adequately represent the population & would lead to incorrect conclusions
about the population
Types of sampling bias:
Convenience bias - when sample is drawn from the most easily accessible part of population
Environment bias - when conditions have changed from the time sample was drawn to the
time sample was used to draw conclusions
Measurement bias - when measurement system is not precise enough
Non-response bias - when respondents do not participate as a sample (survey conditions)

Collect Fresh Data MAIC
D
Fresh data should be collected to ensure that the latest process trend is studied
Historical data may have measurement errors which would be validated in next step of DMAIC
Sometimes, process may generically improve as compared to the ‘Define’ phase

due to increased attention from the owners / error in sampling
In such cases, Champion / sponsors may review the project, address the ‘discipline’ issues &
decide whether project needs to collect another sample or gets abandoned here

How Big a Sample? MAIC
D
Business criteria to select a sample size include cost, time & effort
Statistical criteria include the accuracy of the sample representing the population
Higher the sample size, better the accuracy of the information about the population
parameters ( µ & σ )
Z1 with n = 25
Z2 with n = 16
Z3 with n = 4
There must be a balance between the business & statistical criteria

Determinants of Sample Size - Continuous Data
MAIC
D
The sample size is determined by answering 3 questions
How much variation is present in the population? (σ)
In what interval does the true population mean need to be estimated? ( ± )
How much representation error is allowed in the sample? (α)
Sample size formula for Continuous Data:
2
Z 1 – (α / 2) * σ
n=

Example 1 - Continuous Data
MAIC
D
Let’s take a normal pack of cards. It has got 52 cards. Average of all cards is 7 & standard
deviation is 3.78. Now if I want to take a sample of few cards & want their mean to be within ± 2, i.e.
5 & 9, how many cards should I take?
= 2
σ = 3.78
2
Assume α = 0.05
Z 97.5 * 3.78
n=
2
From Appendix 1, Z 97.5 = 1.96
So, sample size n = [ (1.96 * 3.78) / 2 ]
2
= 14
That means 95% of the samples with size 14 will have its mean between 5 & 9

Example 2 - Continuous Data
MAIC
D
We know that the population standard deviation of runs scored by a cricket player is
25. Now, we want to collect a sample that can estimate the career average within
± 5 runs tolerance with 99% confidence. What should be the sample size?
= 5
2
σ = 25
Z 99.5 * 25
α = 0.01 n=
5
From Appendix 1, Z 99.5 = 2.58
So, sample size n = [ (2.58 * 25) / 5 ]

2
= 166.4
Round off to next higher integer = 167

MAIC
D
Standard Sample Size Formula - Continuous Data
Usually, value of α is taken as 5%
Z 97.5 = 1.96
Thus, standardized sample size formula can be written as
2
1.96 * σ
n= for Continuous Data

Standard Sample Size Formula - Discrete Data D MAIC
Extending the same logic, we can find out the sample size required while dealing with discrete
population
If the average population proportion non-defective is at ‘p’, population standard deviation can be
calculated as
σ = p ( 1 – p)
1.96
n= p ( 1 – p) for Discrete Data
Where = Tolerance allowed on either side of the population proportion average in %

1.1 Map the detailed as-is process
1.2 Define a unit
1.3 Define specifications & defect
1.4 Understand data characteristics
1.5 Find opportunities for error
1.6 Design sampling plan

D MAIC
Step 2
Validate
Measurement System
for ‘Y’

DMAIC Steps

D
 Step 2 Validate Measurement System for ‘Y’ M
Step 4
Step 5
A
Step 7
Step 8
I
Step 10
Step 11
C
Introduction - MEASURE
A robust measurement system forms the basis of any Six Sigma project
A measurement system has two characteristics
Step 1 of DMAIC
Design of the measurement system
Accuracy of the measurement system


Step 2 of DMAIC
Measurement system for ‘Y’ indicates that this step deals with the accuracy of
defect measurement

2.1 Perform GRR study
2.2 Analyze results

2.1 Perform GRR Study

Count the Occurrence of letter ‘I’ in the Paragraph
A country preacher was walking the back-road near a church. He became thirsty so
decided to stop at a little cottage and ask for something to drink. The lady of the house
invited him in and in addition to something to drink, she served him a bowl of soup by the
fire. There was a small pig running around the kitchen. The pig was constantly running up
to the visitor and giving him a great deal of attention. The visiting pastor commented that
he had never seen a pig this friendly. The housewife replied: "Ah, he's not that friendly.
Actually, that's his bowl you're using!"
Explain- Repeatability and Reproducibility for this example

Effect of Measurement Error
Measurement
Variation
LSL USL
Measured Sigma
of the
Process

Objectives of a Measurement Study
Obtain information about the type of measurement variation associated with the
measurement system
Establish criteria to accept and release new measuring equipment
Compare measuring one method against another
Form basis for evaluating a method suspected of being deficient
Resolve measurement system variation

Measurement Analysis
An evaluation of the measurement system MUST be undertaken to ensure effective

analysis of any subsequent data generated for a given process/product characteristic
Observed Value = True Value +/- Measurement Error
Measurement error is a statistical term meaning the net effect of all sources of
measurement variability that cause an observed value to deviate from the true value
True Variability = Process Variability +/- Measurement Variability
Both process and measurement variability must be evaluated and improved together
If we work on process variability first and our measurement variability is large, we can
never conclude that the improvement made was significant, or correct

Types of Measurement Errors
Measurement System Bias (Accuracy) - by Calibration Study
µ total = µ process + µ measurement
Measurement System Variation (Precision) - by GRR Study
σ2 total = σ2 process + σ2 measurement

Accuracy & Precision
Measurement System
Variation
Accurate Precise
Accuracy Stability Linearity Repeatability Reproducibility
Calibration GRR Study

Gage Reproducibility
Gage Reproducibility is the variation in the average of measurements made by

different operators using the same gage when measuring identical characteristics of
the same parts
Operator 2
Operator 1
Reproducibility

Gage Repeatability
Gage Repeatability is the variation in measurements obtained when one operator

uses the same gage for measuring the identical characteristics of the same parts
Repeatability

Components of GRR Study
Difference leads
to Reproducibility
1 2 1 2 1 2
Trial
Reading 3 4 3 4 3 4
#1
5 6 5 6 5 6
Difference leads to
Repeatability
Six Parts / Conditions
1 2 1 2 1 2
Trial
Reading 3 4 3 4 3 4
#2
5 6 5 6 5 6
Operator Operator Operator

A B C

Key Concepts
Gage ‘Repeatability’ & ‘Reproducibility’ studies are referred to as GRR studies
GRR studies should be performed over the range of expected observations
Actual equipments should be used for the GRR studies
Already written procedures should be followed
It should be business as usual
Measurement variability should be presented “as-is”, not as it was designed to be
After GRR, measurement variability is separated into causal components,

prioritized & targeted for action

Data Collection
Usually 3 operators
Usually 10 units to measure
General sampling techniques should be used to represent the population
Each unit is to be measured 2-3 times by each operator (Number of trials)
Gage should have been calibrated properly
Resolution should have been ensured
First operator should measure all units in random order
Same order should be maintained for all other operators
Repeat for each trial

Methods of Performing GRR Studies
X-R (Xbar R) Method
Typically used in automobile industry
Extreme values affect the method
Short method & Long method
Short method does not measure operator & equipment variability separately
Long method measures operator & equipment variability separately, but does not measure
combined effect
ANOVA Method
Measures operator & equipment variability separately with combined effect as well that
better defines causality
More effective when extreme values are present

X-R – Short Method
Only requires 2 Operators & 5 Parts

Tool
GRR – Xbar-R -- Short Method

Tolerance is ‘USL – LSL’ as defined in Step 1

X-R – Short Method

X-R – Long Method
Short method gives total measurement system variation, not separating the
equipment & operator variation (repeatability & reproducibility) Tool
GRR – Xbar-R -- Long Method

For improvement, we must know the real source of variation so as to focus efforts
Long method may not need the ‘tolerance’ value as an input
However, time, resource & cost constraints may need to be looked into
We shall use Minitab for this exercise

X-R – Long Method - Example

X-R – Long Method - Example
Minitab gives the following output:
Source Variance StdDev 5.15*Sigma
Total Gage R&R 7229.94 85.0291 437.900

Repeatability 7229.94 85.0291 437.900
Reproducibility 0.00 0.0000 0.000
Part-to-Part 2026.05 45.0116 231.810
Total Variation 9255.99 96.2081 495.471
Source %Contribution %Study Var
Total Gage R&R 78.111 88.380

Repeatability 78.111 88.380 Number of distinct categories = 1
Reproducibility 0.000 0.000
Part-to-Part 21.889 46.786
Total Variation 100.000 100.000

ANOVA Method
Long method separates the equipment & operator variation, but does not elaborate
Tool 12
on combined effect. ANOVA takes care of this issue
ANOVA uses the ‘standard deviation’ instead of ‘range’
GRR – ANOVA Method

ANOVA also may not need the ‘tolerance’ value as an input
However, time, resource & cost constraints may need to be looked into
We shall use Minitab for this exercise

ANOVA Method - Example
We can use the same data in the previous example with ANOVA method.
Two-Way ANOVA Table With Interaction
Source DF SS MS F P
Part 2 38990 19495.2 2.90650 0.16616

Operator 2 529 264.3 0.03940 0.96173
Operator*Part 4 26830 6707.4 0.90185 0.48352
Repeatability 18 133873 7437.4
Total 26 200222

ANOVA Method - Example
Source VarComp StdDev 5.15*Sigma
Total Gage R&R 7304.7 85.4673 440.157

Repeatability 7304.7 85.4673 440.157
Reproducibility 0.0 0.0000 0.000
Operator 0.0 0.0000 0.000
Part-To-Part 1354.5 36.8036 189.538
Total Variation 8659.2 93.0547 479.232
Source %Contribution %Study Var
Total Gage R&R 84.36 91.85

Repeatability 84.36 91.85
Reproducibility 0.00 0.00
Operator 0.00 0.00
Part-To-Part 15.64 39.55
Total Variation 100.00 100.00

GRR for Discrete Data
Xbar-R & ANOVA methods apply to continuous data only
For discrete data, a relatively higher accuracy is desired
Usually, discrete data GRR is measured against the ‘true’ value
Minimum sample size should be between 15-20

2.2 Analyze Results

Continuous Data
GRR as a % of Tolerance
If GRR as % of tolerance is less than 10% - excellent measurement system
If GRR as % of tolerance is between 10% to 30% - acceptable measurement system
However, discretion may be needed depending upon application of the process / equipment
If GRR as % of tolerance is above 30% - unacceptable measurement system
You should not proceed to next DMAIC step. Simplify process / explore root cause

Continuous Data
GRR as a % of Contribution to Variation / Number of Distinct Categories
If GRR as % of contribution is about 1/10th of the total variation - Acceptable
If GRR as % of contribution is about 1/3rd of the part-to-part variation - Acceptable
If number of distinct categories is >= 4 - Acceptable
If none of the above criteria is met, do not proceed to the next step

Tollgate - Measure
Detailed As-is Process
Units, Specifications & Defects
Number of OFE’s, if discrete data
GRR of the Measurement System
Action plan, if GRR is not acceptable
Reduction of GRR to acceptable level

Control
Improve
ANALYZE
Measure
Define

DM AIC
Step 3
Establish
Process
Baseline

DMAIC Steps

D
 Step 3 Establish Process Baseline
Step 4
Step 5
A
Step 7
Step 8
I
Step 10
Step 11
C
3.1 Collect Data
3.2 Compute Process Sigma Multiple
3.21 Process Sigma Multiple for Discrete Data
3.22 Process Sigma Multiple for Continuous Data
3.3 Estimate population parameters from sample statistics

3.2 Compute Process Sigma Multiple

Six Sigma Metrics
Customer Specifications
Discrete Data Continuous Data
Defects
Location
Defects per Unit
(DPU) Spread
Defects per Million Opportunities Shape

(DPMO)
Process Sigma Multiple - Z

3.21 Process Sigma Multiple for Discrete Data

Defects per Unit
Number of defects found at any check-point

DPU =
Number of units processed at that check-point
It is the universal measure of quality
It may be calculated at a single point or added up for multiple points
It is a count of all defects present in a unit, not of how critical each defect is
DPU may be greater than 1
DPU is an internal measurement, because customers see ‘defectives’

Defects per Million Opportunities
Recollect the concept of opportunities in Measure – Step 1
Opportunities help to compare dissimilar processes by normalizing the complexity
Number of defects found at any check-point

DPO =
Number of units processed Number of opportunities
at that check-point * per unit to create a defect
DPMO = DPO * 1000,000
A multiplier of 1000,000 is used to enable higher sigma measurements
Each DPMO value is unique to a process sigma multiple (Z)

DPMO to Z Conversion Table

Z to DPMO Conversion Table

Z Calculation Worksheet – Discrete Data
Tool
Z Calculation – Discrete Data

Yield = (1 – DPO)

Normal Distribution

Introduction to Normal Distribution
Developed by astronomer Karl Gauss

Tool
Most prominently used distribution in statistics
Normal Distribution
Applicability to many situations where given the population knowledge, we need to
predict the sample behavior
It comes close to fitting the actual frequency distribution of many phenomena
Human characteristics such as weights, heights & IQ’s
Physical process outputs such as yields

Normal Distribution with Figure 3.01

Mean =100
Standard Deviation = 10 1 unit
of
standard
deviation
-  + 
70 80 90 100 110 120 130
It’s a Probability Distribution, illustrated as N ( µ, σ )

Simply put, a probability distribution is a theoretical frequency distribution
Higher frequency of values around the mean & lesser & lesser at values away from mean
Continuous & symmetrical
Tails asymptotic to X-axis
Bell shaped
Total area under the Normal curve = 1

Figure 3.02
-  + 
µ
- 1σ + 1σ
68.26%
- 2σ + 2σ
95.46%
- 3σ + 3σ
99.73%
- 4σ + 4σ
99.9937%
- 5σ + 5σ
99.99943%
- 6σ + 6σ
99.999998%

Standard Normal Distribution
-  + 
µ1 µ2 µ3 -3 -2 -1 0 +1 +2 +3
Instead of dealing with a family of normal distributions with varying means & standard
deviations, a standard normal curve standardizes all the distributions with a single curve
that has a mean of 0 & standard deviation of 1
It’s illustrated as N ~ ( 0,1 ), i.e. mean = 0 & standard deviation = 1

Concept of Z Value
To standardize different measurement units; such as, inches, meters, grams; a

standard Z variable is used
Y - µ
Z=
σ
Where Y= Value of the data point we are concerned with
µ= Mean of the data points
σ= Standard Deviation of the data points
Z= Number of standard deviations between Y & the mean (µ)
Z value is unique for each probability within the normal distribution
It helps in finding probabilities of data points at anywhere within the distribution

3.22 Process Sigma Multiple for Continuous Data

What is A Six Sigma Process?
Target Specification Limit
Higher 66807 Defects

Probability Per Million
of Failure Opportunities
1
3 ’s
Higher this
number,
Lower the
chance of Much Lower

3.4 Defects
producing a Probability Per Million
defect of Failure Opportunities
1
6 ’s

Concept of Process Shift
Short term capability
Sample at Time 3
Sample at Time 2
Shift
Sample at Time 1
Long term performance
LSL T USL

Key Concepts
Short term capability (ZST)
It is the capability or the potential performance of the process, in control at any point of time
Long term performance (ZLT)
It is the actual performance of the process over time
Process shift (ZST – ZLT)
It reflects how well a process is controlled, usually a factor of 1.5 is used

Concept of Process Shift
Over time, a typical process will shift by approximately 1.5 standard deviations
In other words, long term variation is typically 1.5 standard deviations more than the
short term variation
This difference is called the Sigma shift, which is an indicator of process control
This shift could be due to different operators, raw material, wear & tear, time, etc.
Standard deviation of the sample is called as the long term deviation.
DPMO is always calculated on the long term standard deviation
Discrete data Z values as studied in the previous session have been adjusted for shift

ZST, ZLT & DPMO
DPMO indicates long-term performance (Z LT)
Long term performance adjusted by a factor of 1.5 gives short term capability (Z ST)
Short term capability is the sigma multiple of the process (Z ST = Z LT + 1.5)

Process Variations
Common Cause Special Cause
White Noise Black Noise
Un-assignable Cause Assignable Cause
Random Systematic

Process Variations
Short term variation (Common Cause)
Variance inherent in the process (natural variation)
Small number of samples, each sample collected in a short interval

Common cause variation is captured
Common causes can not be identified & corrected (process re-design would be needed)
Long term variation (Common + Special Cause)
Added variation due to factors external to the usual process (abnormal variation)
Large data collected over time
Special causes (different operators, raw material, wear & tear) lead to increase in variation
Special causes need to be identified & corrected for improvement
Long term variation is always greater than the short term variation

Process Capability (CP)
Process Capability USL - LSL

CP =
6 SST
ZST = 3 CP
LSL USL
CP is a measure of short term process capability
CP relates the process short term variation with the customer specification limits
It does not take into account how centered data is
For a six sigma process, CP = 2 (ZST = 3 * 2 = 6)
That means this process can fit 12 standard deviations between USL & LSL

Limitations of Process Capability (CP)
20 140 20 140
LSL USL LSL USL
SST = 10 CP = 2 ZST = 6 SST = 10 CP = 2 ZST = 6
Even though almost 40% of the data is outside specification limits in graph 2, it is
still a Six Sigma capable process, however, not performing to its potential
CP can not be used for one-sided specifications
To address these issues, another measure, called CPK is used

Capability Index (CPK)
USL - Y Y - LSL
CPU = CPL =
3 SST 3 SST
CPK = Minimum ( CPU , CPL ) ZST = 3 CPK
For one sided specifications
CPK is a measure of actual short term process performance
It considers the data centering & forces the mean to be between the specifications
CPK enables ZST computation for one sided specifications
CP > = CPK
If CP is >>> CPK, process is capable but not performing up to its potential

Performance Index (PPK)
USL - Y Y - LSL
PPU = PPL =
3 SLT 3 SLT
PPK = Minimum ( PPU , PPL ) ZLT = 3 PPK
PPK is a measure of actual long term process performance
It is similar to CPK except that it uses the long term standard deviation
PPK enables ZLT computation for both one sided & two sided specifications
Difference between PPK & CP indicates the shift in the process

Key Concepts
Capability is an internal measurement of the process behavior
Performance is an external view of the process behavior
Processes studied in actual projects may not have a shift of 1.5
Due to limitations of multiple shift factors & CP, process sigma multiple calculations
for continuous data start from PPK
ZST
Data PPK ZLT

LSL, USL,
Mean, Variance DPMO

Visualizing the Continuous Data
Below is the histogram drawn on the time taken in pizza delivery
15
10
Frequency
5
35 40 45 50 55 60
Time taken in delivering pizza

Using Minitab for Continuous Data Z Calculation
Process Capability Analysis for time taken in pizza delivery

Process Data
ST USL
USL 60.0000
Target *
LT
LSL *
Mean 47.4667
Sample N 30
StDev (ST) 6.21507
StDev (LT) 6.44895
Potential (ST) Capability

Cp *
CPU 0.67
CPL *
Cpk 0.67
Cpm *
30 40 50 60 70
Overall (LT) Capability Observed Performance Expected ST Performance Expected LT Performance
Pp * PPM < LSL * PPM < LSL * PPM < LSL *
PPU 0.65 PPM > USL 33333.33 PPM > USL 21868.43 PPM > USL 25979.72
PPL * PPM Total 33333.33 PPM Total 21868.43 PPM Total 25979.72
Ppk 0.65

Key Concepts
There is hardly any shift in the process
Observed performance indicates the DPMO calculation if the data is treated as

discrete (i.e. > 60 & <= 60 )
Expected DPMO values are different from the observed DPMO because expected
values are calculated as per the fitted normal probability distribution
Sigma multiple of the process is simply = ( 3 * PPK ) + 1.5
DPMO of the process is 25979, as per the expected long term performance

Defect Classification
You may require classifying defects in terms of where they occur
For example, you can classify late pizza deliveries in terms of outlets from where
the late deliveries were made to increase focus

Appendix 1 – Normal Distribution Table
Z
Area Below +ZLT

Appendix 1 – Normal Distribution Table (contd.)
Z
Area Below +ZLT

DM AIC
Step 4
Define
Performance
Goals

DMAIC Steps

D
 Step 4 Define Performance Goals
A
Step 5 Identify Variation Sources

Step 7
Step 8
I
Step 10
Step 11
C
4.2 Establish Process Improvement Goal

Key Concepts
If the existing process has to be improved, then the improvement goal should be
chosen only after proving statistically that it can be achieved only due to change, &
not by noise
If the improvement is possible even otherwise, discipline issues need to be

addressed
Verify if the targeted performance is really performing at a higher level
However, project teams must not get over-ambitious, entitlement study should
be completed

Introduction to Hypothesis Testing
Suppose team management wants to see if Indian cricket team’s performance Tool
has improved after the have recruited a new coach. Is there an
improvement that can be proven statistically?
What does the management need to do? It basically needs to make an assumption
Test of Hypothesis
about the efficiencies of the two coaches A & B, & test it for significance
Making such an assumption is called Forming a Hypothesis
Forming hypothesis is common around us
If I study for 8 hours a day, I can secure 90% marks

If we drop the price of the TV by 10%, we can increase sales by 25%

Null Hypothesis
When a person is being prosecuted for a crime, the judge hears the proceedings
assuming that the person has committed no crime
The job of the prosecutor is to prove his assumption wrong
In other words, the person is non-guilty till proven otherwise, i.e. status quo
Assuming status quo is Null Hypothesis
Considering the previous example, null hypothesis is that the two coaches have the
same efficiency (i.e. no difference in efficiencies till proven otherwise)

Alternative Hypothesis
Alternative hypothesis challenges the null hypothesis
If null hypothesis is proven wrong, alternative hypothesis must be right
The prosecutor believes in the alternative hypothesis & gives proofs to substantiate it
Considering the previous example, alternative hypothesis is that coach B has a higher
efficiency than coach A (i.e. there is a difference in efficiencies)

Using Hypothesis Testing for Goal Setting
One of the key steps in setting an improvement goal is to prove that the targeted /
benchmarked performance is truly performing at a higher level (different population)
In other words, a project team can set a target hypothesizing that this would be
really an improved performance level
Team must test this hypothesis statistically, otherwise it may end up setting either
too easy or too stiff a target

Hypothesis Testing Roadmap
Basic determinants of accepting or rejecting a hypothesis remains same, however,

various tests are used depending upon the type of data
Hypothesis
testing
Discrete Continuous
data data
Variance Mean
c2 test Comparison
of two
Comparison
of many
F test
σ unknown t test ANOVA
σ known Z test

4.3 Review Financial Benefits & Revise
Improvement Target if Required

Key Concepts
Sometimes, at this step, project teams may set aggressive targets to justify the
project selection
However, one must understand that such steps may result in abandoning the
project half-way
Worst still, it sets a bad precedence & de-motivates the team members
Knowledge about the true behavior of the process is limited at the time of project
selection & hence, benefits expected may be grossly approximated
After setting the improvement target, project team must arrive at an accurate
estimate of benefits with Finance & take the sign-off from Champion to proceed
Sometimes, project teams may need to re-visit step 4.2 depending upon the
Champion’s expectations

DM AIC
Step 5
Identify
Variation
Sources

DMAIC Steps

D

Step 4
Step 5
A
Step 7
Step 8
I
Step 10
Step 11
C
Means & Ends of a Process
We have studied the transfer function in step 4
Y = f ( X1, X2, X3……Xn )
Y X
Dependent Independent
Process Output Process Input / Step
Effect Cause
Symptom Problem
Monitor Control

Means & Ends of a Process
So far, we have focused on the process output ‘Y’
Unit
Specifications
Defects
Baseline
Target
Variation in ‘Y’ is due to ‘X’
In other words, defect observed in output ‘Y’ is due to some ‘X’ not being controlled
In this step, we identify the factors that contribute to variation in the process output ‘Y’
Another objective of the step is to separate the vital few X’s from trivial many

Generating & Prioritizing X’s
Identifying & prioritizing X’s could be done using both non-statistical & statistical tools
Identification tools include (non-statistical basis)

Experience of process doers
Brainstorming & Multi-voting Focus on identifying problem areas
Process mapping
Fishbone
Focus on identifying potential root causes
Prioritization tools include

Pareto Diagram
Correlation (for a single X)
Regression (for multiple X’s)
Chi-square test Focus on prioritizing potential root causes
ANOVA
Failure Mode & Effect Analysis (FMEA)

5.1 Identify potential X’s that contribute to variation in ’Y’
5.2 Separate vital few X’s from trivial many for further screening

Fishbone Example
CRM malfunction Error in Listening

No
Lack of training maintenance
Printer error on listening skills Problem with
CRM not saving the instrument
data properly Rough
handling
Wrong Customer
Backbone Details on Invoice
No credit card verification
Foul order
Problem area Big Bone Medium Bone Effect

5.2 Separate Vital Few from Trivial
Many for Further Screening

Pareto Diagram
Frequencies of root causes for reaching office late

70 100
60
80
50
Percent
Count
60
40
30 40
20
20
10
0 0
me ady y i ng
o n ti re l at
e j am ead ait
ng ot up affi
c ot r sw
o mi es
n
oke Tr stn Bu
t c oth W kf a
no Cl ea
Bu
s Br
Count 25 18 15 6 5 2
Percent 35.2 25.4 21.1 8.5 7.0 2.8
Cum % 35.2 60.6 81.7 90.1 97.2 100.0

Pareto Diagram
Pareto diagram can be used even in step 3 when we establish the process baseline
If a process has multiple defect definitions, project teams can use the Pareto to see
where to focus first for defect reduction, & set improvement targets accordingly
Even for single defect definitions, if there are multiple defect categories (different
products, different geographies, etc.), Pareto could be useful
For example, if late delivery for products is the defect definition, one can use the Pareto to
see if the frequency of late delivery is higher in product A, B or C, & focus accordingly

Key Concepts
So far, we have not used any statistical tool to prioritize X’s.
Depending upon the data characteristics of Y & X, we can choose the appropriate
tool
Correlation
Continuous & ANOVA
Regression
Y Identify
opportunities for
Discrete converting ‘Y’ Chi-square
into a
continuous one
Continuous Discrete
X

Correlation – Association between Variables
If we want to associate ‘Y’ with a single ‘X’, we can use correlation

Tool
Correlation is about predicting the movement in values in ‘Y’ when ‘X’ changes
Statistical significance of that movement is denoted by correlation coefficient ‘r’
‘r’ is always between –1 & +1
Correlation
Positive value of ‘r’ means direction of movement in both variables is same
Negative value of ‘r’ means direction of movement in both variables is inverse
Zero value of ‘r’ means no correlation between the two variables
Higher the value of ‘r’, stronger the correlation between ‘Y’ & ‘X‘
An ‘r’ value of > + 0.85 or < - 0.85 indicates a strong correlation

Correlation Levels
r = 0.05 r = 0.50
6 6
4 4
2 2
0 0
0 6 12 0 6 12
8 10
6 8
6
4
4
2 2
0 0
0 6 12 0 6 12
r = 0.95 r = – 0.95
Correlations of 0.5 or below are hard to see
Correlation measures the linear association between the output (Y) and one
input variable (X) only

Regression
While correlation tells us only about the direction of movement, it does not
throw much light on degree of movement in one variable with respect to Tool
movement in another
Regression of ‘Y’ on ‘X’ results in a transfer function equation that can be used to
predict the value of ‘Y’ for given values of ‘X’
Y = f(X)
Regression
‘Y’ can be regressed on one or more X’s simultaneously
Simple linear regression is for one X
Multiple linear regression is for more than one X’s
Regression by subsets is to choose the best model when there are many X’s
Polynomial regression is to explore the curvilinear relationship between variables

Key Concepts
Even though we get a transfer function Y = f(X) as an output of the regression in

step 5, this is not the correct transfer function to control ‘Y’
Main thrust of regression in this step is only to discover whether a substantial

relationship exists between ‘Y’ & a particular ‘X’, i.e., whether it is a vital ‘X’ or not
The correct definition of transfer function will be arrived at in step 6 in Improve

phase when we perform experiments only on vital X’s

Simple Linear Regression (SLR)
A simple linear regression equation is nothing but a fitted linear equation between
‘Y’ & ‘X’ that looks as follows:
Y = A + BX + C
where Y= Dependent variable / output / response
X= Independent variable / input / predictor
A= Intercept of fitted line on Y axis
B= Regression coefficient / Slope of the fitted line
C= Error in the model

Least Squares method in SLR
If ‘Y’ & ‘X’ are not perfectly linear (r = ± 1), there could be several lines that could be
fitted
Y Y
Error
Error Error
X Y-intercept X
Minitab fits the line which has the least value of errors squared & added

SLR Example
Suppose R&D department of a company wishes to predict the relationship between

amount spent on R&D & annual profit generated. It collects the following data for
last few years & wants to give a R&D budget for this year based on the targeted
company profits. Company has targeted a profit of INR 32 Million this year.
X Y

SLR Example
Minitab gives the following output
Regression Analysis
The regression equation is
Annual Profit = 20.0 + 2.00 * R&D expenditure

Predictor Coef StDev T P
Constant 20.000 2.646 7.56 0.002
R&D expenditure 2.0000 0.4583 4.36 0.012
S = 3.240 R-Sq = 82.6% R-Sq(adj) = 78.3%
Analysis of Variance
Source DF SS MS F P
Regression 1 200.00 200.00 19.05 0.012
Residual Error 4 42.00 10.50
Total 5 242.00
For a targeted profit of INR 32 Million, R&D could be budgeted for INR 6 Million

Interpreting Results
P-value of 0.012 for R&D expenditure denotes that expenditure is statistically

significant to have a non-zero regression coefficient
R-Sq = 82.6% = r2. This is called the coefficient of determination
Also, R-Sq = SSRegression / SSTotal
The R-Squared value is the proportion of variability in the Y variable accounted for by
the predictors. In other words, 82.6% of variation in ‘Y’ is explained by ‘X’ in the fitted
model

Interpreting Results
If you add another variable X2 to the model, value of R2 will increase even though
X2 may not add any value to the model. Minitab adjusts this with R2ADJ value so that
dummy variables are accounted for, if added
R2 values of 70% & above denote a good relationship between ‘Y’ & ’X’ &
that the respective ‘X’ should be further studied
ANOVA table shows the P-value for the regression model which is less than 0.05
indicating that fitted model is good enough

Multiple Linear Regression (MLR)
Multiple regression regresses ‘Y’ on more than one X’s simultaneously
It is similar to Two-way ANOVA we have discussed in step 4, except for the difference
that X’s used in ANOVA were discrete
The approach is similar & a linear multiple regression equation looks as follows:
Y = A + B1X1 + B2X2 + C
where Y = Dependent variable / output / response
X1 = First independent variable / input / predictor

X2 = Second independent variable / input / predictor
A = Intercept of fitted line on Y axis
B1 & B2 = Regression coefficients / Slopes of the fitted plane on two axes
C = Error in the model

Points to Remember in Regression
Don’t extrapolate beyond the range
Project teams get tempted to extrapolate the results beyond the range of collected data. In
one of the previous examples, higher R&D expenditures may not see same increase in
profits, & hence, the regression equation between profit & R&D expense may change
Range 1 Range 2
Profit
R&D Expense

Don’t assume causation

Regression equation denotes a relationship only. This in no way means that a change in one
variable causes change in another. If number of schools & incidents of crime in a city go up
together, there may be a relationship, but no causation. The increase in both factors could be
due to third factor – population.
In other words, both of them may be dependent variables themselves
Number
of schools
Population Incidents
of Crime
Incidents Number of Schools

of Crime
Population

Don’t choose X’s that are correlated
Regression assumes that all X’s are mutually independent variables. If one of the X’s
depends on another X, it may result in a good regression model (ANOVA P-value), but P-
values of the regression coefficients may be insignificant
That means a model with ‘Y’ on any of these X’s will be good, but not when both the X’s
are included in the model

ANOVA
ANOVA is used to short-list potential discrete X’s for a continuous ‘Y’

Tool
We can use one-way ANOVA & see the variation in ‘Y’ for one ‘X’ at a time
We can use two-way ANOVA for more than one X
ANOVA
ANOVA Example
A restaurant puts great emphasis on customer satisfaction. For some weeks, the
ratings seemed to suffer & the manager tried to identify the factors that could be
causing this. He chooses one of the potential factors as ‘team’ that serves the
customer. Team 1 takes care of the Chinese cuisine while team 2 serves the Indian
food. Is the type of team a potential source of variation in satisfaction ratings?

ANOVA Example
One-way Analysis of Variance

Analysis of Variance for Ratings
Source DF SS MS F P
C2 1 8.030 8.030 9.34 0.005
Error 30 25.780 0.859
Total 31 33.810
Individual 95% CIs For Mean
Based on Pooled StDev
Level N Mean StDev --------+---------+---------+-------
1 16 3.0312 0.9300 (-------*------)
2 16 4.0331 0.9240 (-------*-------)
--------+---------+---------+-------
Pooled StDev = 0.9270 3.00 3.60 4.20

Points to Remember in ANOVA
As in regression, do not assume cause & effect in ANOVA also
In the previous example, the root cause may be that Chinese & Indian foods are
prepared by two different chefs, satisfaction ratings for two teams are just an
indicator of the variation. It may not mean that team 2 is more efficient than team 1
When you use multiple X’s in ANOVA, they have to be mutually independent

Failure Mode & Effect Analysis (FMEA)
FMEA is a simple tool to prioritize the failure modes & actions

Tool
By understanding why and how we fail, we can plan for success
It works on the belief that proactiveness saves time
We shall focus on Process FMEA (Design FMEA is used in designing products)
FMEA
FMEA Concept & Output
Effect
Severity
Failure Risk Action

Process / Product
Mode Cause Priority Plan
Characteristics Number
Occurrence
Control
Detectability
Risk Priority Number (RPN) = S*O*D
Severity, occurrence & detectability are measured on a scale of 1-9

FMEA Concept & Output
Process / Product characteristics - Purpose of the product or Process
Failure Mode - How can the product / process fail to function?
Effects - Which effects are most severe to customer?
Causes - Which causes are most likely to occur?
Controls - Ability for current controls to detect causes?
RPN - Which high risk cause we work on first?
Action Plan - Recommended actions & responsibilities

Scales of ‘SOD’ - Severity
Severity is an assessment of the seriousness of the effects of the failure mode on the
customer
Five typical different ratings are used:
1 Minor: Customer will probably not notice failure
3 Low: Customer will notice slight deterioration in product / process
5 Moderate: Customer will notice performance deterioration in product / process
7 High: High degree of customer dissatisfaction and unusable product
9 Very High: Potential failure mode affects safe operation and/or non-compliance
with regulations

Scales of ‘SOD’ - Occurrence
Occurrence is the probability that a specific cause will result in the particular failure
mode
1 Remote: Unlikely cause
3 Low: Occasional occurrence of failures due to this cause
5 Moderate: Relatively few occurrences of failures due to this cause
7 High: Repeated occurrence of failures due to this cause
9 Very High: Failure is almost every time due to this cause

Scales of ‘SOD’ - Detectability
Detectability is the probability that a particular cause will be detected
1 Very High: Detected almost every time
3 High: Good chance of detection
5 Moderate: May be detected by chance
7 Low: Low chance of detection
9 Very Low: No chance of detection

FMEA Table

Key Concepts
Even though, project teams may identify potential X’s using Brainstorming / Fishbone,
& use Regression / ANOVA / Chi-square to prioritize potential X’s, they may still end
up with X’s that explain the variation in ‘Y’, but do not really cause that variation
However, they may just be the real causes
The real output of this step is to short-list potential X’s that may have a causal
relationship with ‘Y’, because a relationship between ‘Y’ & ‘X’ is a necessary but not
sufficient condition for cause & effect
If an ‘X’ does not explain variation in ‘Y’, it should not be explored further, it is one of
those trivial many X’s project team would have identified
A good job done in this step reduces the work in further steps
We check for causation in the next step of Analyze module through experimentation
techniques

Tollgate - Analyze
Baseline Process Sigma multiple / DPMO / Mean / Variance
Target Process Sigma multiple / DPMO / Mean / Variance with statistical

significance
Prioritized list of X’s that contribute to variation in ‘Y’

Control
IMPROVE
Analyze
Measure
Define

DMA C I
Step 6
Explore
Potential
Causes

DMAIC Steps

D
Step 4
Step 5
A
 Step 6 Explore Potential Causes
Step 7
Step 8
I
Step 10
Step 11
C
DMA C I
Step 7
Establish
Variable
Relationship

DMAIC Steps

D
Step 4
Step 5
A
 Step 7
Step 8
I
Step 10
Step 11
C
Means & Ends of a Process IC
DMA
We have studied the transfer function in steps 4 & 5
Y = f ( X1, X2, X3……Xn )
Y X
Dependent Independent
Process Output Process Input / Step
Effect Cause
Symptom Problem
Monitor Control

IC
DMA
Quantitatively Segregate the Vital Few X’s

Creating a List of X’s DMA IC
Brainstorming
FMEA
x1 x7 = 38%
x2 x6 = 27% Vital X’s
x3 x2 = 12%
x4
x5 Exploration of x9 = 4%
x6 the y-x x10= 4%
x7 relationship x5 = 2%
x8 x1 Trivial X’s σ error
x9 x3
x10 x4
x11 x8 = 13%
x12 x11
x12
Fishbone

Design of Experiments (DOE) DMA IC
Definition
Experimental Design is a structured proactive process for investigating the Tool

relationship between input and output factors. Multiple input factors are considered and
controlled simultaneously to ensure that the effects on the (multiple) output responses are
Design of experiments (DOE)

causal and statistically significant
Origin
1920’s with Sir R Fisher
Has an agriculture based nomenclature, e.g. treatments

Design of Experiments (DOE) DMA IC
Strategy of DOE
Define the Problem
Establish the Objective
Select the Response (Y)
Select the Factors (X’s)
Choose the Factor Levels
Select the Experimental Design
Collect the Data
Analyze the Data
Draw Conclusions
Run Additional Experiments, if necessary
Achieve the Objective
Experimental design is more than just analyzing data. It is a structured process for
achieving an objective

Key Concepts IC
DMA
Process experience must be used to design the experiment

Factors
Operating range
Number of levels
Repetition of experiments

DOE Nomenclature IC
DMA
DOE
Experiment Design
Factors
Response
Design
Level

DOE Nomenclature IC
DMA
Experiment Design
The formal plan for conducting the experiment is called the “experiment design” (also the
“experiment pattern”)
It includes the choices of the responses, factors, levels, blocks, and treatments and the
use
of certain tools called planned grouping, randomization, replication
Factors
A factor is one of the controlled or uncontrolled variables whose influence upon the
response
is being studied in the experiment. Factors are also known as the X’s
A factor may be quantitative, e.g., temperature in degrees, time seconds
A factor may also be qualitative, e.g., different machines, different operator, clean or not
clean

DOE Nomenclature DMAIC
Response
The measured characteristic used to quantify the result of a combination of factors at given
levels. The response will be one of the Y’s
Design (Layout)
Complete specification of experimental test runs including blocking, randomization,
replications, repetitions, and assignment of factor-level combinations to experimental units
Level
The “levels” of a factor are the values of the factor being examined in the experiment. For
quantitative factors, each chosen value becomes a level, e.g., if the experiment is to be
conducted at two different temperatures, then the factor temperature has two “levels”. In a
qualitative factor, the single factor “cleanliness” has two levels: clean and not clean.

Pitfalls of Traditional Intuitive Approach IC
DMA
Sometimes, project teams conduct one-at-a-time-factor experiments to study the

effects, using the Simple linear or polynomial regression
This approach is changing one factor at a time, keeping the other factors constant
Consider an example of a car where its fuel efficiency is dependent on the speed
(Km / Hr) & coolant level (ml / Liter) for the engine
Efficiency = f (Speed, Coolant)

Pitfalls of Traditional Intuitive Approach DMA IC
First, experiment runs are made by keeping the speed constant at 155 & varying
the coolant level
Speed Coolant Efficiency
155 0.5 45%

155 1.0 65%
155 1.5 77%
155 2.0 71%
155 2.5 48%
80
75
70
Optimum is slightly to the right of 1.5,
Efficiency
65
60 hence natural conclusion is to

55 increase the coolant level for the next experiment
50
45
40
0 0.5 1 1.5 2 2.5 3
Coolant

Pitfalls of Traditional Intuitive Approach IC
DMA
Now, experiment runs are made by keeping the coolant constant at 1.67 & varying
the speed
Speed Coolant Efficiency
140 1.67 59%

150 1.67 74%
160 1.67 78%
170 1.67 73%
180 1.67 66%
80
75
70
Still about 78% efficiency,
Efficiency
65
60 what is the optimum level then?
55
50
45
40
130 140 150 160 170 180 190
Speed

Traditional Vs. Planned Experiments IC
DMA
One-at-a-time Planned
• No guarantee of finding optimum • Optimum is methodically sought,

statistically verified and documented
• Interaction of variables can
• Interactions of variables are
produce incorrect conclusions incorporated into the design
•Time-consuming and inefficient • Efficient and effective
Interaction is defined as the effect of one factor ‘X1’ on the ‘Y’ being dependent on which level of
another factor ‘X2’ is chosen
For example, al low speeds, fuel efficiency may not get affected by low coolant level. But, at
high speeds, if coolant is low, fuel efficiency may come down drastically
Thus, understanding of combined effects of factors is very important

Pitfalls of Experiments IC
DMA
No objective
A well-defined experimental objective is similar to a good problem statement. It is developed
using standard problem solving tools such as problem statement, cause and effect diagrams
and root cause analysis. An experiment in Step 6 should not encounter this difficulty if the
first five steps of the MAIC have been performed correctly
Process not stabilized

One of the purposes of the trial run phase of experimentation is making sure that the process
and the measurement system are in some kind of control. Ideally, both the process and the
measurements should be in statistical control as measured by a functioning SPC system. At
the very least, process settings should be reasonably reproducible and the variability of the
measurements should be known
Too long experiment

Things around you might change while you are conducting your experiment

Types of Experiment Designs IC
DMA
Screening
& Optimization
Characterization
Full Factorial
Multi-level Experiments
2K Factorial
Composite Designs
Fractional Factorial

Key Concepts IC
DMA
In many process development and manufacturing applications, the number of

potential X’s (factors) is large. Screening (process characterization) is used reduce
the number of X’s by identifying the key X’s or process conditions that affect product
quality. This reduction allows you to focus process improvement efforts on the few
really important X’s, or the “vital few.” Screening may also suggest the “best” or
optimal settings for these factors, and indicate whether or not curvature exists in the
responses. Optimization experiments can then be done to determine the best
settings and define the nature of the curvature.
In industry, two-level full and fractional factorial designs are often used to “screen”
for the really important factors that influence process output measures or product
quality. These designs are useful for fitting first-order models (which detect linear
effects), and can provide information on the existence of second-order effects
(curvature) when the design includes center points.

Key Concepts DMA IC
Screening experiments can be used to quantitatively separate the vital few X’s from
the trivial many X’s. These types of experiments are also known as exploratory
experiments
We have done some amount of screening in analyze phase, albeit non-

experimentally
The second phase of screening experiments might be to form a mathematical

equation relating the y to the x. This step is called characterization
Optimization experiments seek the level(s) of the vital few X’s that will optimize the
performance of the y, with respect to targeting, variability reduction, or both

The Simplest Design – Two Level & Two Factor IC
DMA
High
Graphical Illustration B
Low
Low A High
Trial A B
1 - -
Tabular Illustration
2 + -
3 - +
4 + +

Two Level & Three Factor Design IC
DMA
High
Graphical Illustration B
High
Low
Low High C
A Low
Trial A B C
1 - - -
2 + - -
3 - + -
4 + + -
Tabular Illustration 5 - - +
6 + - +
7 - + +
8 + + +

Balance & Orthogonality IC
DMA
HIGH
(H, L) (H, H)
(+1, -1) (+1, +1)
x1 x1 x2
Factor X1
(L, L) (L, H)
(-1, -1) (-1, +1)
1 x2
LOW
Factor X2
LOW HIGH
Balanced Σ X i = 0 for each factor sum

This feature helps to simplify the analysis
Orthogonal Σ X i X j = 0 for all pairs

This feature ensures the effects are independent

Balance & Orthogonality IC
DMA
If we add all first numbers ( ΣXi ) in the parentheses, sum is ZERO, ditto for second
numbers - balance
If we multiply all first & second numbers in the parentheses & add all up ( ΣXiXj ),
sum is ZERO - orthogonal
A balanced & orthogonal design provides both bits of information – effect of

individual factors, effects of interactions
Without these properties, we lose the ease of analyzing
These properties are desirable but not necessary

Some DOE Settings in Designs IC
DMA
Repetition
This is running the experiment twice on each trial combination, without changing the setting,
i.e. no other run in between
Replication
This is running the experiment twice on each trial combination, but with a change of setting,
i.e. some other run in between
Replicates should be used in 22 &23 designs since number of trials is less
Blocking
A blocking variable is a factor whose levels are used in the experiment, but, effect on
response is not studied
Randomization
Runs are made in random order as opposed to a standard order to avoid lurking variables
that change over time

Key Concepts DMA IC
Repetition & Replication address the issue of experimental error by taking multiple
readings on one setting
Often in a process, there are factors which may have an effect on the response but
are either unknown, uncontrollable, or of no interest to the experimenter. If ignored,
these factors can confound the results and produce erroneous conclusions. With
proper experimental planning, these types of factors can be accounted for with a
minimal effect on the number of trials performed. Randomization addresses the
issue of unknown factors
Nuisance factors that can be classified can be eliminated using a blocked design.
For example, an experiment may be carried out over several days with large
variations in temperature and humidity, or data may be collected in different plants,
or by different technicians. Observations collected under the same experimental
conditions are said to be in the same block. In above case, 1 plant could be 1 block
& even though it is similar to a factor level, its effect is not studied in the experiment

Steps in DOE Analysis IC
DMA
1. Perform the analysis of variance (ANOVA)
2. Look at the p-values for the terms in the model. p-values should be generally less
than 0.05 for significance
3. Watch out for interaction terms being significant
4. Look at effects visually – main effects plots & interactions plots
5. Form two lists, one for significant factors / interactions (vital few) and one for non-
significant factors / interactions (trivial many)
6. Estimate the prediction equation (transfer function) to assess the magnitude and
direction of change in ‘Y’ as a function of ‘X’
7. Use the results from the analysis and develop another design, if necessary

Creating Designs in Minitab DMA IC
Consider a washing machine whose cleaning efficiency is being studied. Three

factors have been identified that affect the extent of cleaning:
Temperature of water (W)
Time elapsed in running the machine (T)
Concentration of cleaning agent (C)
2 levels are chosen for each of the factors
W Warm (-1) Hot (+1)
T Shorter (-1) Longer (+1)
C Lower (-1) Higher (+1)
Response being measured is left out ‘dirt’ content in the clothes measured through a
standard evaporation procedure

Considering a full factorial 23 design, following is the design matrix:
Verify that it is a balanced & orthogonal design matrix
The design matrix can be constructed in Minitab itself
Each trial can be replicated a number of times to provide an estimate of the error in
the experimental process

Creating Designs in Minitab IC
DMA
Completion of the design of the experiment results in following output:

DMA
Experiment is run as per the randomized run order as suggested by Minitab &
following response is observed
Data for ‘dirt’ is entered in Minitab

DMA
Visualizing the cube structure of 3 factors with two levels
Long (+1)
Time
Higher (+1)
Concentration
Short (-1)
Lower (-1)
Lower (-1) Temp. Higher (+1)

DMA
52 41
50 47
50 41
43 45
58 45
Time
61 43
Concentration
65 44
65 Temp. 42

Following is the Minitab Output:
Fractional Factorial Fit

Estimated Effects and Coefficients for Dirt (coded units)
Term Effect Coef StDev Coef T P
Constant 49.500 0.6903 71.70 0.000
Temp -12.000 -6.000 0.6903 -8.69 0.000

Time -6.750 -3.375 0.6903 -4.89 0.001
Conc 0.250 0.125 0.6903 0.18 0.861
Temp*Time 6.750 3.375 0.6903 4.89 0.001
Temp*Conc 0.750 0.375 0.6903 0.54 0.602
Time*Conc 2.500 1.250 0.6903 1.81 0.108
Temp*Time*Conc -2.500 -1.250 0.6903 -1.81 0.108

DMA
Following is the Minitab Output:
Analysis of Variance for Dirt (coded units)
Source DF Seq SS Adj SS Adj MS F P

Main Effects 3 758.50 758.500 252.833 33.16 0.000
2-Way Interactions 3 209.50 209.500 69.833 9.16 0.006
3-Way Interactions 1 25.00 25.000 25.000 3.28 0.108
Residual Error 8 61.00 61.000 7.625
Pure Error 8 61.00 61.000 7.625
Total 15 1054.00

Interpreting Results IC
DMA
STAT > DOE > FACTORIAL PLOTS > Main effects plot
Main Effects Plot (data means) for Dirt
55.0 55.5
52.5
Dirt
50.0
12 6.75
47.5
45.0
43.5
Temp Time Conc
It’s clear that temperature has the greatest effect, time has a moderate effect &
concentration has the least effect

DMA
It’s easy to conclude that concentration does not have much effect on dirt which
may sound a bit strange. But remember this observation is true only within the
range of values used in the experiment. For some other levels of concentration,
time & temperature, one might see a different result altogether
Therefore, factor levels should be chosen as per the normal operating conditions
because the process is going to be run under those conditions only
Project teams must question those results that defy logic, & try to re-experiment
with different levels

DMA
STAT > DOE > FACTORIAL PLOTS > Interaction plot

Interaction Plot (data means) for Dirt
-1 +1 -1 +1
62
Temp
1 52
-1
42
62
Time
1 52
-1
42
Conc

Interpreting Results DMA IC
Temperature & time have significant interaction because change in response for
different levels of temperature is not same for different levels of time
It’s easy to find an interaction by just looking at the two lines; if they are parallel, no
interaction
That means temperature & concentration have no interaction

Finding the Transfer Function DMA IC
We can now develop the coded prediction equation for the response taking into
account only the significant effects
Since prediction equation starts at the grand average level, all effects are divided by
2
Dirt = 49.5 + -12/2 * Temp + -6.75/2 * Time + 6.75/2 * Temp * Time
Dirt = 49.5 – 6*Temp – 3.375*Time + 3.375*Temp*Time
To improve the cleaning efficiency, dirt should be lower. Hence, higher temperature
should be used. We have seen that at higher temperature, time does not matter, &
hence time could be reduced on a practical basis

Key Concepts DMA IC
Process Improvement Strategies through DOE
Initially, use as many X’s as possible
Screen to find vital few X’s.
Concentrate on vital few X’s to optimize their values such that you achieve desired
response
‘Y’ by using a 2-level optimization design in the next step of DMAIC
Use 3-level optimization designs, if additional desirable improvements can be achieved or

warranted

Key Concepts
Vital few X’s were identified & characterized in step 6. Since Y = f(X) is known from
step 6, the X’s in the transfer function are the vital ones. Step 7 attempts to
establish the levels of these X’s that provide the desired improvement in ‘Y’
Analysis of the transfer function reveals whether an increase in ‘X’ either increases
or decrease the output ‘Y’
For a target value of ‘Y’, the required level of ‘X’ can be determined analytically
Focus of step 7 is on optimization experiments. After choosing the level of ‘X’ (to
give the target value of ‘Y’), the process is checked to affirm the result
Even when a transfer function is well-defined, there are typically still trivial X’s
affecting the process. The aggregate level of their effect should be determined

7.1 Determine optimum levels of Vital Few X’s

7.1 Determine Optimum Levels of Vital
Few X’s

Optimizing ‘Y’ for Discrete Vital X’s IC
DMA
If there are only a few discrete levels of a vital ‘X’, then in previous Step , the best
level would have been identified
Since the optimum level for each of the vital X’s is explicitly identified, it is only
necessary to confirm that the predicted result occurs
Hence, much of optimization techniques we learn in this step , apply to continuous
X’s

Optimizing ‘Y’ for Continuous Vital X’s IC
DMA
In the case of continuous X’s, it may be necessary to explore additional levels of the
vital X’s, in order to gain a better estimate of transfer function f(X)
Either a numerical or graphical depiction of f(X) should provide sufficient information

to estimate which ‘X’ levels optimize ‘Y’
Once the level of each ‘X’ is chosen, it is always a good idea to verify the function.
Confirmation runs should be made to verify that the predicted value of ‘Y’ occurs
The error in the transfer function should be estimated to affirm that the error is
sufficiently small to achieve the goal set . If it is not, then additional X’s may be
necessary to create an enhanced transfer function

Improving the Transfer Function IC
DMA
Screening &
Optimization
Characterization
Once the vital few factors have been identified, a sequential series of experiments
can be used to determine the optimum factor combination
If additional runs are needed to help estimate the transfer function, DOE can be used
again

Points to Remember in Optimization Designs DMA IC
One should know how to use experimental designs to determine the factor settings
which will produce optimal results
One should know common types of experimental designs which use more than two
factor levels
One must understand the advantages/disadvantages of design types
One should be able to analyze the results of optimization experiments and

determine if further improvement is possible and/or desirable

Transfer Function Enhancement Cycle DMA IC
2 level optimization - determining if “better levels” of performance exist

Ascent / Descent Method
Simplex Method
3 level optimization – find the best place to operate when you are already in
the right area
Transfer Function Enhancement
Screening
Performance DOE
goal on ‘Y’
(2K, 2K-P)
2 level optimization 3 level optimization

Two Level Optimization Designs IC
DMA
Consider a cleaning process that has two controllable factors

Time of cleaning & Temperature of water
The response that we are trying to maximize is cleaning efficiency
Design Matrix is as follows with 5 readings on a single center point
Temperature
40.0 41.5
160
40.3
40.5
40.7
40.2
40.6
150
39.3 40.9
Time
30 40

DMA
A single trial at the center point provides insight into the non-linear response
Multiple trials at the center point allows for an estimate of error
Are the current operating conditions optimum? If not, which direction provides
improvement?
‘Analyze Factorial Design’ can be used to see the direction of linear improvement,
(provided we had used Minitab to create the design)

DMA
Screening DOE output is as follows for the cleaning example which signifies that
the fit is significant & the factors are important
Fractional Factorial Fit
Estimated Effects and Coefficients for Yield
Term Effect Coef Std Coef t-value P

Constant 40.4444 0.05729 705.99 0.000
Temp 0.6500 0.3250 0.08593 3.78 0.009
Time 1.5500 0.7750 0.08593 9.02 0.000
Analysis of Variance for Yield
Source DF Seq SS Adj SS Adj MS F P

Main Effects 2 2.82 2.82 1.41 47.82 0.000
Residual Error 6 0.17 0.17 0.02
Curvature 1 0.002 0.002 0.002 0.08 0.791
Lack of Fit 1 0.002 0.002 0.002 0.06 0.821
Pure Error 4 0.170 0.170 0.043
Total 8 3.00

Ascent / Descent Method from Screening DOE DMA IC
The un-coded transfer function equation is
Tool 30
Yield = 40.44 + 0.325*Temp + 0.775*Time
Since effects are significant, we can move ahead
Ascent / Descent Method

Both the coefficients have positive multipliers, direction of improvement in
response is indicated below
Temperature Direction of Ascent

40.0 41.5
160
40.3
40.5
40.7
40.2
40.6
150
39.3 40.9
Time
30 40

Proportional Ascent / Descent IC
DMA
We usually take the center of the design region as the origin point for the path of
steepest ascent. Then steps along the path are chosen proportionally to the signs
and magnitudes of the regression coefficients. Usually, the variable with the largest
regressor coefficient represents the base step and steps for other variables are
then a fraction of this step
Since time has the largest regressor coefficient, it is chosen as the base step size
of 5 (center to face)
Step size for temperature is chosen as
5 * coeff. of temp / coeff. of time = 5 * 0.325 / 0.775 = 2.1

Proportional Ascent / Descent DMA IC
The following subsequent set of trials could now be run:-
Time Temperature
35 155.0
40 157.1
45 159.2
50 161.3
55 163.4
60 165.5
65 167.6
Go as long as you see improvement, but take care of the process noise
Replicates should be taken if the process is assumed to have large noise
If there is a practical barrier (time / cost / skill) on a factor level, keep increasing the
level of another factor

Simplex Method DMA IC
A second 22 experiment could be run with one of the previous corner points
Tool 31
as a base point
The original design is shown as solid lines and this second simplex is shown as
dashed
Temperature
Simplex Method
40.0 41.5
160
40.3
40.5
40.7
40.2
40.6
150
39.3 40.9
Time
30 40
You may choose to replicate / run the experiment again at the (40, 160) setting

Transfer Function Enhancement Cycle DMA IC
Screening
Performance DOE
goal on ‘Y’
(2K, 2K-P)
After the improvement region has been located, we want to find the ideal operating
values
More extensive experiments should be conducted to determine the shape of the

response curve
Since the screening designs / 2 level optimization designs only model linear surfaces,
an augmented design is needed to quantify the curvature of the surface

Enhancing the Transfer Function DMA IC
If there is still too much noise in the process not leading to satisfactory sigma level
of the process as targeted in step 4, most significant trivial X’s from step 5 must be
included in the model & steps 6-7 be repeated
x7 = 38% x7 = 38%
x6 = 27% Vital X’s x6 = 27%
x2 = 12% x2 = 12% Vital X’s
x9 = 4%
x9 = 4% x10= 4%
x10= 4%
x5 = 2% x5 = 2%
x1 Trivial X’s
x1
x3 σerror x3
Trivial X’s Smaller
x4 x4
x8 x8 σerror
x11 x11
x12 x12

Key Concepts IC
DMA
For the purpose of this CTQ, setting the vital X’s to optimize ‘Y’ is the best policy.
Tradeoffs in multiple CTQ’s and Y’s will be explored later
Similarly, since the trivial X’s do not significantly affect this y, these X’s can be set
arbitrarily or with respect to other CTQ’s
It is important to quantify the noise/error in the process once all the X’s are set.
Remember that not all X’s need to be set, as trivial X’s which are uncontrollable will
still vary
If the noise/error does not lead to an acceptable Sigma level, an enhanced cycle
must be performed

Re-visiting Transfer Function Enhancement DMA IC
Screening
Performance DOE
goal on ‘Y’
(2K, 2K-P)

Key Concepts DMA IC
Screening experiments such as 2k factorials and fractional factorials are used to

segregate the X’s that strongly influence the variability and targeting of ‘Y’
For quantitative X’s, a prediction equation can be formed that will provide a direction
of improvement. Sequential screening may be needed using 2-level optimization
designs
Once the vicinity of the optimum settings has been located, 3-level optimization
experiments can be used to quantify the nonlinear effects and compute the best
settings for the X’s
If sufficient improvement has not been made, additional X’s (from the screening
experiment) will need to be added to the pool of vital few

DMA C I
Step 8
Design
Operating
Limits

DMAIC Steps

D
Step 4
Step 5
A
Step 7 Establish Variable Relationship
I
 Step 8 Design Operating Limits

Step 10
Step 11
C
Key Concepts
Step 8 provided the experimental techniques to establish the values of X’s that
produce the best output level of ‘Y’
The best values of X’s are backed off to find a range of values that while not the
‘best’ level, does provide acceptable output levels
A range of ‘X’ values can provide additional flexibility in setting these factor levels
while not adversely affecting the output, specially in the cases of multiple responses

8.1 Determine operating limits of vital few X’s
8.2 Perform resistance analysis

8.1 Determine Operating Limits of Vital
Few X’s

Defining Operating Limits
Y = f (X)
USL
LSL
LOL UOL
X
XL XU
XT
Range of acceptable X values

Optimal X setting

Effect of Vital Few X’s
Y
Target
X3
For one ‘Y’ value, two ‘X’ values could be set
While setting the ‘X’ value, following must be considered

Variability in ‘Y’ measurement (GRR)
USL & LSL for ‘Y’
Strength of calculated X-Y relationship
Interaction of this X with other X’s
Practical limitations of set-points for ‘X’
Variability in ‘X’ measurement (GRR)

Determining the Operating Limits
CASE I: No measurement Variability in ‘Y’
Y = 20 + 2X
USL = 90
T = 60
LSL = 45
X
XLSL XT XUSL

Assume the true relationship between ‘X’ & ‘Y’ is demonstrated by the regression line
Y = 20 + 2X
No model error & variability in ‘Y’ is assumed in this case
Solving the equation: X = 0.50(Y – 20)
For a target of 60 on ‘Y’, XT = 20
Since ‘Y’ has USL & LSL as well, we have a range of operating values for ‘X’
For USL & LSL as 90 & 45 for ‘Y’, we can solve for XUSL = 35 & XLSL = 12.5
Hence, value of ‘X’ should range from 12.5 to 35, with an ideal setting at 20

CASE II: Measurement Variability in ‘Y’
Y = 20 + 2X
USL = 90
T = 60
LSL = 45
X
XL XLOL X XUOL XU
T

As studied in step 2 of DMAIC, output ‘Y’ may have a GRR associated with it
We create a buffer against this error by slightly reducing the operating range for ‘X’
New operating limits are called as lower operating limit (LOL) & upper operating
limit (UOL)
If the GRR standard deviation σM = 2 from GRR study, we create a three sigma
buffer for LSL & USL for ‘Y’ & adjust them to 90 - 6 = 84 & 45 + 6 = 51
Solving for XLOL & XUOL from X = 0.5(y-20)
XLOL = 15.5 XUOL = 32

Hence, cases I & II combine as follows:
Y = 20 + 2X
USL = 90
USLB = 84
T = 60
LSLB = 51
LSL = 45
X
XL XLOL XT XUOL XU
12.5 15.5 20 32 35

Operating Limits for Various X Types
Y
Y
USL
LSL
X X
13 14 15 16 17 18 19 20 21 22 23
Continuous x Discrete Numerical x
Y
USL
LSL
Shift
A B C D
Discrete attribute X

Operating Limits for Multiple Y’s
Y1
X1
Y2
X1
If there is no overlapping region, all CTQ’s can not be fulfilled & project team should
seek help from BB & Champion to prioritize

Should-Be Process Mapping
Once the operating limits are determined, it may be worthwhile to have a re-look at
the process
New operating limits may need some extra steps, some steps to be scrapped, or
some steps to be performed differently

8.2 Perform Resistance Analysis

Understand Resistance Sources
Why bother?
Need sufficient support and involvement from key stakeholders
Critical mass must be won over
Key difference between success and failure
What are we after?
Coalition of committed resources
Identification of potential resistance
Conversion of key influencers

Sources of Resistance
Technical
Habit and inertia
Difficulty in learning new skills
Sunk costs
Lack of skills
Political
Threats to old guard from new guard
Relationships
Power and authority imbalance
Cultural
Selective perception
Locked into old "mindset"
Afraid of letting go

Sources of Resistance
Sources of Resistance Causes of Resistance
What in the world is Six Sigma??

Technical Don’t expect the Sales team to come for training, we
have got targets to meet!!!!
Why don’t you ask Finance department to improve?

Political What have I got to do with a “quality control”
initiative?
But our sales have been improving!!!!!!!

Cultural I already work 12 hours a day!!!!
Remember what happened to ISO???

Influencing Strategies
Technical
Do an alignment test for systems and structures
Provide training and education
Provide coaches, tools, job aids
Run pilot to demonstrate
Political
Do a political map to understand influence patterns
New measures and rewards
Clarify roles and responsibilities – accountabilities
Involve champion
Cultural
Do a cultural audit: what beliefs drive us?
Articulate desired mindset and gaps
Redefine measures and rewards
Make known important core values that remain constant

Tollgate - Improve
Vital Few X’s & their best settings
Operating Limits for each vital ‘X’
SOP’s for new settings & training to operators, if required
Should-be Process Map
Counter-measures for possible resistance sources, if any

Control
Improve
Analyze
Measure
Define

DMAI C
Step 9
Validate
Measurement System
for ‘X’

DMAIC Steps

D
Step 4
Step 5
A
Step 7
Step 8
I
 Step 9 Validate Measurement System for ‘X’
Step 10
Step 11
C
Key Concepts
In step 9, we determined the operating limits of vital few X’s. Step 9 applies the
GRR tools used in step 2 of DMAIC to assess the measurement system variability
associated with these X’s
Operating Limits as determined in step 9 may need to be adjusted depending upon

the GRR

9.2 Analyze results


Buffering the Variability
Y
Measurement Variability in ‘Y’ & Model Error (other X’s)
Y = f (X)
USL
LSL
X
XL XLOL XUOL XU
Measurement Variability in ‘X’

Key Concepts
There are three major sources of variability to buffer against:
Measurement variability on ‘Y’ step 2
Model Error step 6 & 7
Measurement variability on ‘X’ step 9

Key Concepts
Measurement validation could be applied after Step 5. Since it may not be known
at that time which X’s are vital, measurement validation would need to be done on
all X’s
Measurement validation could be applied after Step 6, when the vital X’s are
known. The operating region for the vital X’s might not be known
Measurement validation could be applied after Step 7, but similarly the region of
operation for the X’s may not have been completely determined
Measurement validation could be applied after Step 8, which obviously fits the step
order. This time may be a little late for X’s with high measurement error, since
decisions have been already made without knowledge of this noise

Destructive Measurement System for ‘X’
A destructive Measurement test occurs when the same part/condition cannot be

tested by the same operator a multiple number of times
In this case when the conditions are not repeatable and reproducible, the best
estimate for equipment and appraiser variation comes from when the conditions are
held as homogenous as possible. However, it is confounded with the sample
variation

Difference leads to
Reproducibility & Sample Variability
1 2 7 8 13 14
Trial
Reading 3 4 9 10 15 16
#1
5 6 11 12 17 18
Difference leads to
Repeatability &
36 Total Parts / Conditions
Sample Variability
19 20 25 26 31 32
Trial
Reading 21 22 27 28 33 34
#2
23 24 29 30 35 36
Operator Operator Operator

A B C

The GRR for destructive testing can still use multiple operators and multiple trials,
however it is by definition a must to use multiple samples. Note that there are now
36 unique samples for the GRR study
A comparison of observations 1-6 with 19-24 still gives a sense of repeatability

(equipment variation). However, this variability is now confounded with sample
variability
Likewise, a comparison of observations 1-6, 7-12, and 13-18 yields a variability

estimate on operator and sample
The variability associated with sample is not separable from either equipment or
operator. Thus, one should try to obtain samples which are as homogenous as
possible

9.2 Analyze Results

Key Concepts
Specific rules for GRR on X’s are developed from the operating ranges
In this step, “GRR as a % of tolerance” is also important (tolerance is nothing but

the operating range for the ‘X’)
If the measurement standard deviation is sufficiently large so as to eliminate the

operating range of ‘X’ while creating the buffer, measurement system is inadequate
Thumb rules as studied in step 2 still apply, however, one must test the effect on ‘Y’
& see if it remains within the specification limits

Choosing the Final ‘X’ Set-point
Y = f (X)
USL
LSL XLOL XUOL
X
XL XU
Final Buffered Operating Limits

Choosing the Final ‘X’ Set-point
Operating limits of all vital X’s are buffered to satisfy the ‘Y’. The target ‘X’ value will
give the best DPMO. However, every value of ‘X’ within the final buffered operating
limit will satisfy the DPMO & Sigma multiple requirement

DMAI C
Step 10
Verify
Process
Improvement

DMAIC Steps

D
Step 4
Step 5
A
Step 7
Step 8
I
 Step 10
Step 11
C
Key Concepts
Step 10 emphasizes the principles of statistical process control in sustaining the

improved performance
Step 10 also re-computes the process baseline to ensure that targeted level of
performance has been achieved as promised in step 4

10.1 Develop control mechanism
10.2 Prepare action plan to maintain X’s that are varying away from settings
10.3 Re-compute process baseline & verify against target

10.1 Develop Control Mechanism

Controlling the X’s
So far, we have identified the best settings for each of the vital ‘X’
The key now is to ensure that the X’s don’t vary away from the targeted setting
Process control is a crucial tool in ensuring that this Six Sigma project delivers
lasting benefits
Maintaining X’s at their target level can be done in two ways

Detection (reactive) Why was change not detected?
Prevention (proactive) Why did / would change occur?

Correction Loop
Prevention
Prevention Detection &
Detection
Mistake-Proofing Acceptance Sampling Statistical Process

Control

Mistake-Proofing (Poka-Yoke)
It is a technique to make errors difficult to happen, if not impossible

Tool
Ground Rules of Mistake Proofing
Intelligence does not depend upon hierarchy
Target tasks that require constant alert / vigilance / memory
Mistake-Proofing
Look for cutting unproductive time to foster creativity
Defects are not acceptable, in whatsoever small number
It is best to make it impossible to do it wrong

Why do Errors Occur?
Errors occur due to following reasons
Incorrect procedures (SOP’s)
Too much variation within the process

Too much variation in the process inputs
Operator error
Poor design
How does mistake proofing help?
100% error-proof inspection
Immediate feedback to enable action

Steps in Mistake-Proofing
Brainstorming
Identify Customer complaints
1
Problems Error reports
Rejection analysis
FMEA
Prioritize
2 Cost & Effort estimate
Problems COPQ
Identify Don’t cover-up

3 Don’t treat symptoms
Root-cause Correct errors at source
Find Make it impossible to go wrong

4
Solutions Be creative – think out of the box
Measure Have errors been eliminated?

5
Outcome What is the financial impact?

Mistake-Proofing Techniques
Technique Prevention Detection
A mistake is about A mistake has been

SHUTDOWN to be made made
Defective parts can’t

CONTROL Defects are impossible move to next process
step
Something is about Something has just gone

WARNING to go wrong wrong

Mistake-Proofing Techniques - Examples
Technique Prevention Detection
Cameras that don’t click Air-conditioners trip when

SHUTDOWN if the shutter is on detect slight over-heating
ATM does not accept a Quality check at each

CONTROL card inserted wrongly point in the assembly
Car security system

WARNING alerting the driver that all Smoke detectors
doors are not closed

Advantages of Mistake-Proofing
Easy to accomplish without any formal training
Removes repetitive tasks
Fosters creativity & value-addition
Ensures less defects

If Mistake-Proofing is Not Possible?
Provide guidelines
Check-lists
SOP’s
Templates
Use visuals
Color-codes
Shapes

Class Exercise – 10 Minutes
Each team takes any two of the situations & apply mistake-proofing techniques:
• Car owners complain that they often forget if the fuel tank hole is on the left or right side of the
car.
• A cold drink manufacturer wants to ensure that all bottles are filled with exactly the same
quantity.
• Commuters complain that ‘free left’ is always blocked by the vehicles that have to go straight.
• HR team of a company has found that employees only punch-in & don’t punch-out.
• Bank customers complain that they find it difficult to keep track of cheques issued by them.
• Administration team of a company finds that employees don’t switch-off lights while leaving.
• A FMCG company has found that retailers don’t disburse the freebies to customers as due.
• Credit card customers complain that they end up over-spending on their cards.

Acceptance Sampling
Sometimes it may be needed to know when a particular lot should be rejected or

accepted, given the number of defective X’s found in the sub-group sampled
Tool
In general, the decision is either to accept or reject the lot so as to make sure that the
defective X’s do not result in out of specification ‘Y’. This process is called Acceptance
Acceptance Sampling
Sampling
Acceptance sampling was originally applied by the U.S. military to the testing of bullets
If every bullet was tested in advance, no bullets would be left to ship
If, on the other hand, none were tested, malfunctions might occur in the field of battle,
with potentially disastrous results

Acceptance Sampling
A sample should be picked at random from the lot, and on the basis of information that
was yielded by the sample, a decision should be made regarding the disposition of the
lot (population)
The number of defects found in the sample of size ‘n’ is compared to a predetermined
standard, the critical number of defects ‘c’. If the observed number exceeds the critical
number, the entire lot (population) is rejected
This concept is similar to that of hypothesis testing

Characteristics of Acceptance Sampling
Acceptance sampling is "the middle of the road" approach between no inspection and
100% inspection
A point to remember is that the main purpose of acceptance sampling is to decide

whether or not the lot is likely to be acceptable, not to estimate the quality of the lot
Acceptance sampling is employed when one or more of the following hold:

Testing is destructive
The cost of 100% inspection is very high
100% inspection takes too long
An acceptance sampling plan (ASP) is a sampling scheme and a set of rules for
making decisions. The decision, based on counting the number of defectives in a
sample, can be to accept the lot, reject the lot, or even, to take another sample and
then repeat the decision process

Limitations of Acceptance Sampling
SPC is preferred over Acceptance Sampling because SPC provides real-time

monitoring of the process
On the other hand, Acceptance Sampling ignores the process & focuses exclusively
on the output after it has been produced
Moreover, SPC provides a long-term control mechanism compared to Acceptance

Sampling which provides a short-term view

Statistical Process Control (SPC)
SPC was developed by Walter A. Shewhart in 1924

Tool
Historically, SPC has been used to monitor & control output ‘Y’
In DMAIC, we apply SPC to control X’s (remember ‘Y’ is only monitored)
Statistical Process Control

However, sometimes applying SPC to ‘Y’ can also be beneficial in detecting trends
About SPC
Aids visual monitoring & controlling
Depends heavily on data collection

Foundation of SPC
It forms data into patterns which can be statistically tested and, as a result, leads to
information about the behavior of process output / control variable characteristics
It graphically represents output / control variable performance
It detects assignable causes which affects the central tendency and/or variability of
the cause system
It serves as a probability-based decision making tool
It points out where action can be taken with known degrees of risk and confidence

Benefits of SPC
SPC applied on ‘Y’ SPC applied on ‘X’
Separates special & common Recognizes unintended change in

cause variability process activities quickly
Recognizes unintended change in Enables maintenance within operating

process output quickly limits
Identifies stable zone for calculating Provides useful internal information

process capability for continuous improvement
Provides useful external information

SPC Tools
SPC primarily uses ‘Control Charts’
Upper Control Limit (UCL)

Characteristics
Lower Control Limit (LCL)
Time / Number
‘Process Control’ is inherent to process characteristics as against ‘Process

Capability’
which is measured as per outside targets & specifications

Basics of Control Charts
Control charts are useful for tracking process statistics over time and detecting the
presence of special causes
A process statistic, such as a subgroup mean, individual observation, or weighted

statistic, is plotted versus sample number or time. A “center line” is drawn at the
average of the statistic being plotted for the time being charted. Two other lines—
the upper and lower control limits—are drawn, by default, 3σ above and below the
center line
A process is in control when most of the points fall within the bounds of the control
limits, and the points do not display any nonrandom patterns

Purpose of Control Limits
Special Cause Variation

UCL
Common Cause
Variation
LCL
Special Cause Variation

Purpose of Control Limits
0.135%
UCL = µ + 3σ
99.73%
LCL = µ - 3σ
0.135%
Out of control point
Control Limits define a probabilistic level

of occurrence of an ‘out of control’ point

Steps in Control Charting
1. Select process characteristic to be controlled / monitored
2. Select type of control chart
3. Choose rational sub-groups, if required
4. Determine sample size
5. Establish sampling interval / frequency of sampling
6. Collect initial process data
7. Calculate Center Line and Control Limits
8. Collect ongoing data to evaluate the characteristics
10. Update control limits if necessary
11. Follow SOP’s if an ‘out of control condition’ occurs

Choosing An Appropriate Control Chart
Continuous Data
Individual Data Points Subgroups

Pulling one sample at fixed frequency Taking periodic grouped data
Variability of individual Variability of average

I & MR characteristics over time characteristics over time
when sub-group size X&R
is less than 8
Detection of small
EWMA changes in process Variability of average
characteristics over time
when sub-group size
X&S
is more than 8
EWMA

Choosing An Appropriate Control Chart
Discrete Data
Defectives Defects
Constant Varying Constant Varying

Sub-group Sub-group Sub-group Sub-group
Size Size Size Size
NP P C U
# of units % of units Average # of

# of defects
rejected rejected defects per opportunity

Continuous Control Chart Constants
These constants are used to determine control limits & other process statistics

I & MR Control Chart
Temperature of the cleaning agent is a vital ‘X’ to successfully meet the

requirements of t cleaning efficiency. This is monitored periodically to keep it within
desired operating range.
Data on temperature ( in 0C) is collected over several hours

I & MR Control Chart
I and MR Chart for Temperature
85 1
3.0SL=82.93
Individual Value
75
X=69.07
65
55 -3.0SL=55.20
Subgroup 0 5 10 15
20
3.0SL=17.04
Moving Range
10
R=5.214
0 -3.0SL=0.00

Summary of I & MR Control Chart
LCLX = X - A2 R
UCLX = X + A2 R
LCLMR = D3 R
UCLMR = D4 R
subgroup-size (n) = 1

X & R Control Chart
Let’s take data of the previous example only. Assume that the data on temperature
was collected using three different probes & below table gives three readings per
hour, each for one probe, over 5 hours (5 samples, each of sub-group size 3)

X & R Control Chart
Minitab gives the following output
Xbar/R Chart for Temperature
80
1
3.0SL=77.25
Sample Mean
70 X=69.07
60 -3.0SL=60.88
Subgroup 1 2 3 4 5
20 3.0SL=20.59
Sample Range
10
R=8.000
0 -3.0SL=0.00E+00

Summary of X & R Control Chart
LCLX = X - A2 R
UCLX = X + A2 R
LCLR = D3 R
UCLR = D4 R
subgroup-size (n): 1<n<8

Summary of X & S Control Chart
LCLX = X - A3 S
UCLX = X + A3 S
LCLS = D3 S
UCLS = D4 S
subgroup-size (n): n>8

NP Control Chart
NP charts track the number of defectives and detect the presence of special
causes. Usually, it plots defectives coming from a fixed sub-group size that allows
an apple-to-apple comparison while plotting numbers
Each entry in the worksheet column is the number of defectives for one subgroup,
assumed to have come from a binomial distribution with parameters n and p
Process proportion defective, p, is estimated by the overall given sample
proportion. The center line and control limits are then calculated using this value
Center Line = n p
n = Sub-group size
LCLNP = np - 3  n p (1 – p) p = Total defectives / Total units
UCLNP = np + 3  n p (1 – p)

NP Control Chart
Let’s assume that the quality control department checks the quality of finished
goods sampling a batch of 10 items every hour. If items are found out of control
limits consistently in any given day, production process has to be stopped for the
next day. They collect the following data over 24 hours:

NP Control Chart
NP Chart for NP defectives

1
5
3.0SL=4.725
4
Sample Count
2
NP=1.417
1
0 -3.0SL=0.00E+00
0 5 10 15 20 25
Sample Number

P Control Chart
P charts track the proportion of defectives and detect the presence of special
causes. It allows to plot defectives coming from a varying sub-group size because it
doesn’t use numbers as against NP chart
Each entry in the worksheet column is the number of defectives for one subgroup,
assumed to have come from a binomial distribution with parameters n and p
Process proportion defective, p, is estimated by the overall given sample

proportion. The center line and control limits are then calculated using this value
Center Line = p
n = Sub-group size
LCLP = p - 3  p (1 – p) / n p = Total defectives / Total units
UCLP = p + 3  p (1 – p) / n

P Control Chart
Let’s take the same data used in NP chart example.

P Control Chart
P Chart for P defectives

1
0.5
3.0SL=0.4725
0.4
Proportion
0.3
0.2
P=0.1417
0.1
0.0 -3.0SL=0.00E+00
0 5 10 15 20 25
Sample Number
It’s identical to NP chart in the previous example since sub-group size is constant &
data is same

C Control Chart
C charts track the number of defects and detect the presence of special causes.
Usually, it plots defects coming from a fixed sub-group size that allows an apple-to-
apple comparison while plotting numbers
Each entry in the worksheet column is the number of defects for one subgroup,
assumed to have come from a Poisson distribution with parameters λC
Process average number of defects λ is estimated by the given sample data; this
value also forms the center line of the data. control limits are then calculated using
this value
Center Line = λC
LCLC = λC - 3 λC λC = Average number of defects
UCLC = λC + 3 λC

C Control Chart
Let’s assume that the customer service department administers a questionnaire on

employees which has to be answered in ‘yes / no’. There are total 15 questions.
Each question that is answered in a ‘no’ is a defect. These questions form a vital ‘X’
in measuring employee satisfaction.

C Control Chart
C Chart for C defect
5 3.0SL=4.950
4
Sample Count
2
C=1.400
1
0 -3.0SL=0.00E+00
0 5 10 15
Sample Number

U Control Chart
U charts track the number of defects per opportunity and detect the presence of
special causes. Usually, it plots defects coming from a varying sub-group size
Each entry in the worksheet column is the number of defects for one subgroup,
assumed to have come from a Poisson distribution with parameters λ
Process average number of defects per opportunity λU is estimated by the given

sample data; this value also forms the center line of the data. control limits are then
calculated using this value
Center Line = λU
LCLU = λU - 3 λU / n λU = Total defects / total opportunities

n = Sub-group-size
UCLU = λU + 3 λU / n

U Control Chart
Let’s slightly change the data used in example of C chart. Let’s assume that the
customer service department now administers two questionnaires on employees,
one with 10 & another with 20 questions, i.e. sub-group size varies. They have to
be answered in ‘yes / no’. Each question that is answered in a ‘no’ is a defect.

U Control Chart
U Chart for U defect
0.4 3.0SL=0.3886
0.3
Sample Count
0.2
0.1 U=0.09545
0.0 -3.0SL=0.00E+00
0 5 10 15
Sample Number

Points to Remember in Control Charts
Do not apply SPC tools to processes that are known to be out on control
Do not compare control limits with specification limits
Do not ignore ‘out-of-control’ signals if your ‘Y’ is meeting the specifications & ‘X’ is
meeting the operating limits
Do not plot them for the sake for it, act on it
Ensure that observations are independent of each other
It’s quite possible that ‘X’ is under control, but ‘Y’ is out-of-spec’s

Key Concepts
Success of control charts (specially Xbar-R chart) depends upon the proper
selection of sub-groups
Always remember the thumb-rule - selection of sub-groups should allow the

maximum possible opportunity for sub-group to sub-group variation & minimum
possible opportunity for within sub-group variation
All control charts allow specifying the historical mean & standard deviation, as
applicable. A good practice in control charting is to fix the center line & control limits
by using these parameters so that fresh control limits are not calculated for each
sample
However, control limits must be re-calculated if data displays a clearly different trend & the
reason for this change (new machine / skill / technology / material) is known & desirable

Fixing the Control Limits & Center Line
Let’s use the same example that was used in Xbar-R chart. We measured the
temperature using three probes & calculated the centerline & control limits using
that sample data itself. Suppose we know that population mean & standard
deviation are 70 & 5 respectively. Now we can force the control chart to perform
tests as per these limits for the current sample

Fixing the Control Limits & Center Line
Control chart would now look like this, check the new center line & control limits
Control limits are at: µ ± 3σ / n0.5
Xbar/R Chart for Temperature
80
3.0SL=78.66
Sample Mean
70 X=70.00
-3.0SL=61.34
60
Subgroup 1 2 3 4 5
3.0SL=21.79
20
Sample Range
10
R=8.465
0 -3.0SL=0.00E+00

10.2 Prepare an Action Plan to Maintain
X’s That are Varying Away from
Settings

Action Plan for Out-of-Control X’s
Purpose of Action Plan
Define what corrective actions should be taken when ‘X’ is found to be out-of-control
Define what ongoing actions would keep ‘X’ in control

Should be developed by the project team jointly
Documentation of Action Plan
Must outline current control methods
Should identify current measurements used
Points to Remember
Keep the CTQ in mind
Take help from the process flow diagrams developed in previous steps
Pay attention to FMEA output

10.3 Re-compute Process Baseline &
Verify Against Target

New Process Capability
It is important to now re-compute the process baseline in order to

Verify improvement levels
Compare new capability with predicted capability
Re-confirm noise levels
Form new confidence intervals
It’s possible that all vital X’s are under control, but required improvement is not made
If X’s chosen do not functionally relate to ‘Y’

If some vital X’s are missed out
If the optimum region has not been explored completely
If the operating limits are not fixed properly
If considerable measurement error is present in both ‘X’ & ‘Y’
If required improvement is not made, each of the above points should be explored

Tollgate – Step 10
Improved performance sustained for at least one-two months* with all vital X’s
under control
* Black Belts must use their discretion based upon the sample size available to statistically prove the
improvement

DMAI C
Step 11
Implement
Process
Controls

DMAIC Steps

D
Step 4
Step 5
A
Step 7
Step 8
I
Step 10 Verify Process Improvement
C
 Step 11 Implement Process Controls

Key Concepts
Step 11 emphasizes the need to sustain the improvement made so that the process
does not slip back to the original performance

11.1 Develop action plan to sustain improvement
11.2 Sing-off the project closure form

11.1 Develop Action Plan to Sustain
Improvement

Problem Solving Flow
MEASURE ANALYZE IMPROVE CONTROL
Business Statistical Statistical Business

Problem Problem Solution Solution
P R O J E C T
Business Problem: Low production yield

Statistical Problem: Low mean, high deviation
Statistical Solution: Change raw material spec’s
Business Solution: Develop new vendor

Problem Solving Flow
Business problem is discovering customer & business CTQ’s, establishing

performance parameters & validating measurement systems
Statistical problem is determining the baseline, setting a target performance &

identifying variation sources
Statistical solution is discovering vital X’s, their best settings & operating limits
Business solution is institutionalization of the results – control, audit & translation

How to Sustain Process Improvement
Control Plan
Training & Communication
Project handovers with completed project dockets
Database of vital X’s
QFD / VOC
FMEA / Fishbone
Six Sigma Project Audits
Process Owners
SOP’s

11.2 Sign-off The Project Closure Form

Project Closure
All Six Sigma projects must be closed with sign-offs
Tool
Project Closure Template

Tollgate - Control
Improved performance sustained for at least one-two months1 with all vital X’s
under control
Process owner2 identified for sustenance
Project docket handed over to Champion / Black Belt
1 Black Belts must use their discretion based upon the sample size available to statistically prove
the improvement
2 In cases where GB may move out of the present role

DMAIC Summary
IF right customer
CTQ has been
selected
IF it is
translated to a
measurable Customer
internal CTQ
will
IF vital X’s are be
discovered that
functionally relate
satisfied
to ‘Y’
IF operating limits
are set for vital X’s

Six Sigma

Uploaded by

Copyright:

Available Formats

Six Sigma

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Six Sigma

Uploaded by

Copyright:

Available Formats

What is the difference between the old and new concepts of quality according to Six Sigma?

What is the difference between the old and new concepts of quality according to Six Sigma?

What are some examples of organizations that have adopted Six Sigma globally?

What are some examples of organizations that have adopted Six Sigma globally?

D M A I C

Six Sigma Study Material

SAMIR SHAH Slide 1 Six Sigma Module1-

Past concepts of quality focused on “conformance to standards”. This definition

SAMIR SHAH Slide 2 Six Sigma Module1-

SAMIR SHAH Slide 3 Six Sigma Module1-

• Conversion efficiency of materials

• Inadequate resource utilization

• Excessive use of material

• Lost customers / Goodwill

SAMIR SHAH Slide 4 Six Sigma Module1-

Six Sigma is the most important initiative GE has

You can see the “before” & “after” of an organization

SAMIR SHAH Slide 6 Six Sigma Module1-

Why : Business Strategy, World Class Performance

Where : Customer Critical Criterion

What : Defect Reduction

How : Projects, DMAIC, People Development, Tools

Who : Black Belts, Green Belts, Team Members

When : Usually 3/4 months per project

SAMIR SHAH Slide 7 Six Sigma Module1-

Continuos defect reduction in Products & Services

Global acceptance to Quality System

SAMIR SHAH Slide 8 Six Sigma Module1-

SAMIR SHAH Slide 9 Six Sigma Module1-

The term ‘Sigma’ taken from the Greek alphabet, is used

Sigma capability is a metric which indicates how the

SAMIR SHAH Slide 10 Six Sigma Module1-

Six Sigma is an approach to achieve customer satisfaction

Organizations have no choice but to offer the best

SAMIR SHAH Slide 11 Six Sigma Module1-

High 66807 Defects

SAMIR SHAH Slide 12 Six Sigma Module1-

That means a process efficiency of 99.99966%

Practical Solution Statistical Solution

SAMIR SHAH Slide 13 Six Sigma Module1-

Six Sigma Black Belt / Core Team

SAMIR SHAH Slide 14 Six Sigma Module1-

Is the Head of the Business Unit / Division

Approves Green Belt project along with the Sponsor

Reviews projects periodically with Sponsor

SAMIR SHAH Slide 15 Six Sigma Module1-

Sponsor ( Manager / Section Head )

Selects Green Belt project along with the Black Belt

Reviews projects of Green Belt frequently with Black Belt

Provide resources to Green Belt Leader for accomplishment of Target.

SAMIR SHAH Slide 16 Six Sigma Module1-

Black Belt ( GIHL Six Sigma Experts )

Trains others in Six Sigma methodologies & concepts

Conducts programs on Statistical tools

Provides Application Assistance by facilitating team discussions

Helps review projects with Business Unit Head

Facilitation of Green Belt Projects.

SAMIR SHAH Slide 17 Six Sigma Module1-

Green Belt ( Engineers / Supervisors )