Six Sigma

Six Sigma

A scientific and practical method to achieve improvements in a

company

Scientific:
Structured approach.
Assuming quantitative data.
Show me
the money

Practical:
Emphasis on financial result.
Start with the voice of the customer.

Show me
the data

The Six Sigma Evolutionary Timeline

1818: Gauss uses the normal
curve to explore the
mathematics of error analysis for
measurement, probability
analysis, and hypothesis testing.
1736: French
mathematician
Abraham de
Moivre
publishes an
article
introducing the
normal curve.

1924: Walter A. Shewhart

introduces the control chart and
the distinction of special vs.
common cause variation as
contributors to process problems.
1896: Italian sociologist Vilfredo
Alfredo Pareto introduces the
80/20 rule and the Pareto
distribution in Cours dEconomie
Politique.

1949: U. S. DOD issues Military

Procedure MIL-P-1629,
Procedures for Performing a
Failure Mode Effects and
Criticality Analysis.

1960: Kaoru Ishikawa

introduces his now
famous cause-andeffect diagram.

1941: Alex Osborn, head

of BBDO Advertising,
fathers a widely-adopted
set of rules for
brainstorming.

1970s: Dr. Noriaki Kano

introduces his twodimensional quality model
and the three types of
quality.

1986: Bill Smith, a senior

engineer and scientist
introduces the concept of Six
Sigma at Motorola

1995: Jack Welch

launches Six Sigma at
GE.
1994: Larry Bossidy
launches Six Sigma at
Allied Signal.

Where can Six Sigma be applied?

Service
Management

Design

Purchase

Administration

Six Sigma
Methods

Production

IT

Quality
Depart.
HRM

M&S

the Six Sigma quality

certification was
established by the
International Quality
Federation in 1986, to
judge the quality
standards of an
organisation and
Motorola was the first to

The Six Sigma Initiative

integrates these efforts

Knowledge
Management

Six Sigma is. . .

A performance goal, representing 3.4 defects for every

million opportunities .

A series of tools and methods used to improve or design

products, processes, and/or services.

A statistical measure indicating the number of standard

deviations within customer expectations.

A disciplined, fact-based approach to managing a business

and its processes.

A means to promote greater awareness of customer needs,

performance measurement, and business improvement.

Whats in the name?

Sigma is the Greek letter representing the

standard deviation of a population of data.

Sigma is a measure
of variation
(the data spread)

What does variation mean?

20

Variation means that a

process does not produce
the same result (the Y)
every time.
15

10

-5

Some variation will exist in

all processes.
-10

Variation directly affects customer

experiences.

Customers do not feel averages!

Examples of the Sigma Scale

In a world at 3 sigma. . .

In a world at 6 sigma. . .

There are 964 U.S. flight

cancellations per day.

1 U.S. flight is cancelled

every 3 weeks.

The police make 7 false

arrests every 4 minutes.

There are fewer than 4

false arrests per month.

In one hour, 47,283

international long
distance calls are
accidentally
disconnected.

It would take more than

2 years to see the same
number of dropped
international calls.

5-10-15-20 goals of Six

Sigma
5%

reduction in productions costs,

10% growth in sales, 15% growth in
profit and 20% improvement in cashflow.

Basic Statistics

Population is the universe under study. A

population
could consist of all invoices sent to customers, all
customers inquires seeking technical support, or
a
production run
Sample is small group of observations actually
available
Mean is the measure of central tendency of
population
Mode is the most frequently observed value for a
characteristic

Two General Kinds of Data

1. Discrete Data (a.k.a. Attribute Data)
Type . 1: One can count both the number of
occurrences and non-occurrences
Heads / Tails, Yes / No, Pass / Fail
Type . 2: One can only count the occurrences, e.g.
number of scratches on a car hood, number of
employee accidents per month
2. Continuous Data (a.k.a. Variable Data)
It is a measured data and decimal subdivisions are
meaningful, e.g. time to answer the telephone

Two General Kinds of Data

Contd.

Exercise: Which Type of Data is it?

1. Cycle time for a .Credit Check. process
2. Percent cream content in milk bottles
3. Sales Hit Rate (number of sales proposals that were won)
reported each month
4. Number of defects per square yard of cloth
5. Number of employee accidents per month
6. Proportions of orders that were late coming out of sales dept.
in daily sample of 100 orders
7. Percent defective parts in hourly production
8. Number of surface blemishes in four piece sets of coffee cups
9. Length of screws in samples of size ten from production lot

WHAT
WHAT IS
IS THE
THE MEASURE
MEASURE
OF
OF CENTRAL
CENTRAL TENDENCY
TENDENCY OF
OF
A
A SET
SET OF
OF NUMBERS?
NUMBERS?
There

are three ways in which Central

Tendency of Numbers can be measured.

These

are the 3 Ms

MEAN
MEDIAN
MODE
16

MEASURES
MEASURES OF
OF
DISPERSION
DISPERSION
extent of the spread of the

The
values
from the mean value is called Dispersion.

The measures of Dispersions are

Range (R)
Standard Deviation (s)
Variance (s2)
Co-efficient of Variation (CV)

Standard deviation is the most

commonly
used
measure
of
dispersion.
17

Basic Statistics

Conditions for six sigma

The

process should be stable.

Construct control charts for finding
the stability of the process
Six sigma is meaningless if the
process is out of control

Six sigma process

If

USL and LSL is exactly at +or 6

sigma then the process is said to be a
six sigma process provided there is no
shift in mean.

Sigma level is the number of standard deviations

between the process mean and the nearest
specification limit in sigma units. As process standard
deviation goes up, fewer standard deviations will fit
between the mean and the nearest specification
limit, decreasing the sigma number and increasing
the likelihood of items outside specification

What Are Control Limits?

Process
Distribution

40

Returns

35
30
25
20

Stable Process

A control limit defines the bounds of common-cause variation in the process. A control
limit is a tool we use to help us take the right actions.
If all points are between the limits, we assume only common-cause variation is
present
(unless one of the other Signals of a Special Cause is present).
If a point falls outside the limit, you treat it as a special cause /
Otherwise, you do not investigate individual
data points, but instead study the
21
common-cause variation in all data points.

How to Calculate Control

Limits
Centerline = X

UCL = X + 3 Sigma
LCL = X 3 sigma

UCL = Upper Control Limit. LCL = Lower Control Limit

22

Specification Limits vs.

Control Limits
Specification limits
Are

set by the customer, management, or engineering requirements.

Describe what you want a process to achieve.

Control limits

Are calculated
from the
data.
Describe what
the process is
capable of
achieving.

120

UCL

110

Upper
spec

100

Lower
spec

90

LCL

80

70
1

9 11 13 15 17 19 21 23 25 27 29 31 33 35

23

Tolerance Limits vs. Process

Capability
Specification Width

Actual Process Width

Specification Width

Actual Process Width

Process Capability Example

Design
LTL

Specs: Bearing diameter 1.250 +- 0.005 inches

= 1.245 inches UTL = 1.255 inches

The

actual distribution from the process mean =

1.250, s = 0.002
+-

3s limits 1.250 +- 3(0.002) [1.244, 1.256]

Anew

process, std. dev. = 0.00083

Principles of SPC
Process variation distribution checks

Measuring Process
Performance

Customers
want
their
pizza
The
pizza
delivery

delivered fast!

example. . .

Guarantee = 30 minutes or less

What

if we measured performance and found

an average delivery time of 23.5 minutes?
On-time

performance is great, right?

Our customers must be happy with us, right?

How often are we delivering

on time?
Answer: Look at
the variation!

30 min. or less

Managing by the average

x doesnt tell the whole story. The
0
10 the variation
20
30
40 whats happening.
50
average
and
together
show

Reduce Variation to Improve

Performance
How many standard
30 min. or less
deviations can you
fit within
s
customer
expectations?

10

20

30

40

50

Sigma level measures how often we meet (or fail

to meet) the requirement(s) of our customer(s).

PROCESS CAPABILITY.

30

Potential Process Capability

(Cp)
Cp=

specification width/process spread

Cp= (USL-LSL)/6
Cp

should always be greater than 2.0

for a good process which is under
statistical control. For a good process
under statistical control, Cpk should be
greater than 1.5."

Process capability index

The ability of a process to produce output
within specification limits

The

 when

there is a one-sided
specification limit, or when the target
is not specified, Cpk may be more
conveniently calculated as:

Relationship between Cp,

Cpk and Sigma level
If

the process mean is centered, that

is = T, and USL LSL = 6, it is
easy to know that Cp = 1, and the
distance from to the specification
limit is 3. In this case, the sigma
(quality) level becomes 3, and the
relationship between Cp and the
sigma level is

 In

general, the relationship

between Cpk and the sigma level is
Sigma level = 3 Cpk +1.5
= 3 (Cpk + 0.5)
Cpk = Cp 0.5

 Cpk

>= 1.5 (A Highly Capable Process)

: Cpk = 1 to 1.5 (A Barely Capable
Process)
Cpk < 1 (The Process is not Capable)
The indices for process capability are
based on the assumption that the
underlying process distribution is
approximately bell shaded or normal.

 Cp-Measures

the variation.,
Cpk - Measures the center tendency.,
Cpk is always less than CP

Cpu =Cpl=Cpk =Cp Only when the

process is centered about the target
value

Calculation of DPMO

Calculation of DPMO Contd

Managing Up the Sigma

Scale

Sigma
1
2
3
4
5
6

% Good % Bad

DPMO

30.9%

69.1%

691,462

69.1%

30.9%

308,538

93.3%

6.7%

66,807

99.38%

0.62%

6,210

99.977%

0.023%

233

99.9997% 0.00034%

3.4

how to compute DPU, DPO,

DPMO and sigma level
Case

of DPU
Suppose that the pinhole defects in a
coating process have been found in five
units out of 500 units inspected from a
long-term investigation then DPU =
5/500 = 0.01, the probability of zero
defect is
y = eDPU = e0.01 = 0.99005

 and

the corresponding Z value is Z =

2.33. Since the set of data has been
obtained for a long-term, the shortterm sigma
level is Zs = 2.33 + 1.5 = 3.83

Sigma quality level

In practice, we desire that the process mean to be kept at
the target value. However, the process mean during one
time
period is usually different from that of another time period
for
various reasons.
This means that the process mean constantly shifts around
the target value. To address typical maximum shifts of the
process mean, Motorola added the shift value 1.5 to the
process mean. This shift of the mean is used when
computing a process sigma level

DMAIC The Improvement

Methodology
Define MeasureAnalyzeImprove Control
Objective:
DEFINE the
opportunity

Objective:
Objective:
MEASURE current ANALYZE the
performance
root causes of
problems

Objective:
IMPROVE the
process to
eliminate root
causes

Objective:
CONTROL the
process
to sustain the gains.

Key Define Tools:

Cost of Poor
Quality (COPQ)
Voice of the
Stakeholder
(VOS)
Project Charter
As-Is Process
Map(s)
Primary Metric
(Y)

Key Measure
Tools:
Critical to Quality
Requirements
(CTQs)
Sample Plan
Capability
Analysis
Failure Modes
and Effect
Analysis (FMEA)

Key Improve
Tools:
Solution
Selection Matrix
To-Be Process
Map(s)

Key Control
Tools:
Control Charts
Contingency
and/or Action
Plan(s)

Key Analyze
Tools:
Histograms,
Boxplots, MultiVari Charts, etc.
Hypothesis Tests
Regression
Analysis

Define DMAIC Project

What is the project?

Projec
t
Chart
er

$
Cost of
Poor
Quality

S ta k e h o ld e r s

Voice of the
Stakeholder

Six Sigma

What is the problem? The problem is the

Output (a Y in a math equation Y=f(x1,x2,x3)
etc).
What is the cost of this problem
Who are the stake holders / decision makers
Align resources and expectations

Define As-Is Process

How does our existing process work?

Move-It! Courier Package Handling

Process
Courier

Mail Clerk

In-SortClerk

In-SortSupervisor DistanceFeeClerk

WeightFeeClerk

Accounts
ReceivableClerk

Accounts
Supervisor

Out-SortClerk

Out-SortSupervisor

Observ e package
weight (1 or 2) on
back of package

Look up
appropriate
Weight Fee and
write in top middle
box on package
back
Add Distance &
Weight Fees
together and write
in top right box on
package back

Accounting

Take packages
f rom Weight Fee
Clerk Outbox to
A/R Clerk Inbox.

Take packages
f rom A/R Clerk
Outbox to
Accounts
Superv isorInbox.
Take packages
f rom Accounts
Superv isor
Outbox to OutSort Clerk Inbox.

Delivery

Finalizing

Take packages
f rom Out-Sort
Clerk Outbox to
Out-Sort
Superv isorInbox.

Does EVERYONE
agree how the
current process
works?
Define the Non
Value Add steps

Circle Total Fee

and Draw Arrow
f rom total to
sender code

Write Total Fee

f rom package in
appropriate
Sender column on
Accts. Supv .s log

Draw 5-point Star

in upper right
corner of package
f ront

Sort packages in
order of Sender
Code bef ore
placing in outbox

Add up Total # of
Packages and
Total Fees f rom
log and create
client inv oice

Observ e sender
and receiv er
codes and make
entry in Out-Sort
Superv isors log

Deliv erPackages
to customers
according to N, S,
E, W route

Deliv er inv oiceto

client

Submit log to
General Manager
at end of round

Submit log to
General Manager
at conclusion of
round.

Submit log to
General Manager
at end of round

Define Customer
Requirements

of the
Key Customer
Critical to
What are the Voice
CTQs?
What
motivates
the
Customer
Issue
Quality
customer?

Listening
Posts

Industry
Intel

SECONDARY RESEARCH
Market
Data

Customer
Service

Industry
Benchmarking

Customer
Correspondence

PRIMARY RESEARCH
Surve
ysSurve
ys

OTM

Focus Groups

Observations

Measure Baselines and

Capability

What
is our
Sample some
data /current
not all data level of performance?
Descriptive Statistics

Variable: 2003 Output

Current Process actuals

measured against the Customer
expectation
What is the chance that we will
succeed
at this level every time?
Pareto Chart for Txfr Defects

Anderson-Darling Normality Test

A-Squared:
P-Value:

10

20

30

40

50

95% Confidence Interval for Mu

Mean
StDev
Variance
Skewness
Kurtosis
N

23.1692
10.2152
104.349
0.238483
0.240771
100

Minimum
1st Quartile
Median
3rd Quartile
Maximum

0.2156
16.4134
23.1475
29.6100
55.2907

95% Confidence Interval for Mu

21.1423
19.5

100

21.5

22.5

23.5

24.5

40
20
0

Defect
79
79.0
79.0

17
17.0
96.0

4
4.0
100.0

25.5

26.5

25.1961

95% Confidence Interval for Sigma

8.9690

95% Confidence Interval for Median

Percent

Count

60
50

Count
Percent
Cum %

20.5

100
80

0.211
0.854

11.8667

95% Confidence Interval for Median

19.7313

26.0572

Measure Failures and

Risks

Where does our process fail and why?

Subjective opinion mapped into an objective risk profile number

X1
X2
X3
X4
etc

Analyze Potential Root

Causes
Ishikawa
Diagram
(Fishbone)

What affects our process?

Six Sigma

y = f (x1, x2, x3 . . . xn)

Analyze Validated Root

Causes
Pareto Chart for Txfr Defects

100

100

What are the key root causes?

Count

60
40

Percent

80

50

20

E x p e r im e n ta l D e s ig n

Defect
Count
Percent
Cum %

79
79.0
79.0

Data
Stratificatio
n

17
17.0
96.0

4
4.0
100.0

Regression
Analysis

Pareto Chart for Amt Defects

100

15

60

10

40

Percent

Count

80

Process
Simulation

5
20
0

Defect
Count
Percent
Cum %

12
70.6
70.6

3
17.6
88.2

2
11.8
100.0

Six Sigma

y = f (x1, x2, x3 . . . xn)

Critical Xs

 Address

Improve Potential
Solutions

the we
causes,
not
How can
address
thethe
root causes we
symptoms.
identified?
Evaluate

Clarify

Generate

y = f (x1, x2,
xCritical
xn)
3 . . .Xs

Divergent |
Convergent

Decision

Improve Solution
Selection
Solution Selection

How do we choose theMatrix

best solution?
Quality

Time

Solution

Sigma

Cost

Six Sigma
Implementation
Bad
Good

Solution
Right
Wrong

Nice
Try

Nice
Idea

Solution
Implementa
tion Plan

Time

CBA

Other

Score

Control Sustainable
Benefits

Some variation
normal and
How
do weishold
theOKgains of our new
How High and Low can an X go yet not materially impact
process?
the Y

Pre-plan approach for control exceptions

35
UCL=33.48

Individual Value

25

Mean=24.35

LCL=15.21

15
0

10

20

Observation Number

30

Barriers to implementation
Barrier #1: Engineers and managers are not interested in mathematical statistics
Barrier #2: Statisticians have problems communicating with managers and
engineers
Barrier #3: Non-statisticians experience statistical anxiety which has to be
minimized before learning can take place
Barrier # 4: Statistical methods need to be matched to management style and
organizational culture

Training
Course

level
Belts
Overall vision
Champion
Most comprehensive master black
belt
Comprehensive
Black belt
Median
Green
Basic
white

Seven Steps for Six Sigma

Introduction

Step 1: Set up the long-term vision of Six Sigma.

Step 2: Identify core processes and key customers.
Step 3: Define customer requirements and key process
variables.
Step 4: Measure current process performance.
Step 5: Improve process performance.
Step 6: Design/redesign process if necessary.
Step 7: Expand and integrate the Six Sigma system.

Performance standards
First Time Yield in multiple stage process

Number
Numberof
ofprocesses
processes
11
10
10
100
100
500
500
1000
1000
2000
2000
2955
2955

3
3

4
4

5
5

6
6

93.32
93.32 99.379
99.379 99.9767
99.9767 99.99966
99.99966
50.09
99.77
99.9966
50.09 93.96
93.96
99.77
99.9966
0.1
53.64
97.70
99.966
0.1
53.64
97.70
99.966
00
4.44
89.02
99.83
4.44
89.02
99.83
00
0.2
79.24
99.66
0.2
79.24
99.66
00
00
62.75
99.32
62.75
99.32
00
00
50.27
99.0
50.27
99.0

Financial Aspects
Benefits of 6approach w.r.t. financials

Six Sigma Companies