History and Latest Development

of
Robust Engineering
(Taguchi Methods)
ASQ NORTH JERSEY SECTION 0304
SPRING QUALITY CONFERENCE
"QUALITY FOREVER
Hanover Marriott, Whippany, NJ
APRIL 05, 2012

Dr. Genichi Taguchi

Shin Taguchi
CTO, ASI Consulting Group
President, American Supplier Institute, Inc.
[email protected]
2011 Copyright by ASI Consulting Group, LLC All rights reserved

History of Taguchi Method

T Methods

10s

Standardized S/N for Non-linear Ideal Function

Automotive
Robust Behavior Testing Mahalanobis Taguchi System
Hall of Fame
00s Robust Assessment Dynamic S/N Application to other Systems
1993
Robust Technology Noise Compounding
QES
90s Robust Design
Energy Thinking
Operating Window
80s

Dynamic S/N Application to Mechanical System

Imported to
USA

Non-dynamic S/N Tolerance Design

On-line Quality Engineering
Quality Loss Function
70s Dynamic S/N for Measurement
Inner Array
60s
50s

Parameter Design
Design
Of
Experiment

2007 ASI Consulting Group, LLC

Focus on Work
Linear Graph
ANOVA
OA

Outer Array

2-Step Optimization

Deming Prize
India

Robustness
By Control x Noise

Customer Wants and Needs

The Kano Model
Satisfied

Didnt do
it at all

Did it
very
well
BASIC
NEEDS

Dissatisfied
2007 ASI Consulting Group, LLC

Basic Needs and

Performance Needs

Focus of
Robust Engineering
&
Quality Loss Function

Dr. Taguchis work on Caramel Candy

- Morinaga Seika Co. 1948

y = Caramel Hardness

FUNCTION!!

Robustness !!
Before

When cold, can not even chew this

caramel, and when hot, it melts in pocket.

Robust Caramel !!
It just melts nicely
in your mouth.

After

0 deg.

10 deg. 20 deg. 30 deg. 40 deg.

NOISE FACTOR!!

2007 ASI Consulting Group, LLC

Temperature (C)
4

Dr. Taguchi with

Electrical Communication Labs in 1950s
Development of Cross Bar Switching System
ECL (Electrical Communication Labs.) vs. Bell Labs.

Budget # People # Years

Result
50
5
7
Not finished
AT&T Bell Labs
NT&T ECL

Superior

Lease
Sell

ECL

Contractors
Mfg.
companies

2007 ASI Consulting Group, LLC

Design

Users
Tel Co.
End
users

Leasing by ECL has

100% Warranty
- 40 Years for
Exchanger
-15 Years for
Tel. sets

During six years of

development, ECL has optimized
over 3000 Design Control Factors
for Robustness, i.e. looking for
robustness by studying
interactions between
Control and Noise.
5

Genichi Taguchi
W. E. Deming
Yuin Wu
1979 Basement of Dr. Demings House

2007 ASI Consulting Group, LLC

Dr. Taguchi at the Bell Laboratories

In 1981, Dr. Taguchi offered a lecture at the Bell Laboratories.
Dr. Taguchi asked them to bring the most difficult problem that Bell Labs was facing. .
They challenged Dr. Taguchi with 256k IC
Photolithography Window Size problem.

Epoch Making Case Study at The Bell Labs

Photolithography was used to create 150,000 windows (holes)
on one chip with specification of: 3.00 0.25m

Yield was only 33%

Failure Modes: No window, Window Too big,
Window Too small, Shape Variability, etc.

How would you attack this problem?

2007 ASI Consulting Group, LLC

.......
.......
.......
IC Chip

.......
.......
.......
.......

256k Chip Photolithography Window Size

Bell Laboratories in 1981
Response
y = Window Size
(Nominal-the-Best)

Photolithography

Noise Factors
Chip to Chip
Within Chip
P1

Control Factors
Plasma Etching Time
Exposure Time
Spin Speed
Bake Time
Aperture, Etc

Apply
2-step
Optimization

Step-2

Baseline
Performance

Step-1

P2

Q1

Q2

Three 2-level factors

Six 3-level factors

Q3
Q4

2007 ASI Consulting Group, LLC

Q5

L18(23

36)

Target
( 3.0m )

Yield: 33%

87%

Published in:
Bell Systems Technical Journal, May 1983

Typical Taguchi Method Case Study from 80s.

Chrysler had Porosity problem with sheet molding compound
for automotive interior. Porosity is void that is a defect resulting in poor quality.
Because of Porosity, production direct-run yield was only 77%,
costing over $1million annually for Scrap & Rework.
Response
y Number of Porosity
(Smaller-the-Better)
Control Factors
Forming Pressure, Forming
Temperature, Cycle Time,
Pattern Design, Pretreatment,
Viscosity, Glass Type,
Amount of Additive, etc.

2007 ASI Consulting Group, LLC

Deign of Experiment was

conducted, to look for an optimum
combination of control factors to
minimize
Number of Porosity.
It was confirmed the optimum
process resulted in Number of
Porosity reduced to nearly zero.

Yield was improved to 98% and the

annual scrap & rework cost reduced
by more than $1 million.
ROI was more than 32X times.
9

Fire-fighting Proactive Prevention

Sheet Molded Compound
Process Improvement
Chrysler, 1985
Response
y # of Porosity
STB
Objective
Minimize: and y

Control Factors
Forming Pressure,
Temperature, Cycle Time,
Pattern Design, Under
Coating, Weight,
Glass Type, etc.

Control Factors were

assigned to an L16 with
selected two-factor
interactions.
It was confirmed that the
optimum condition resulted
in just about no porosity.

Direct Run Yield was

improved from 74% to 96%.
Cost reduction due to
reduced rework was
$900,000/Yr.

Not objective of
Taguchi Methods.

Best in Class
Good Company
Typical Company

I have a big problem

with my son, Shin.
He helps his clients to
fire fight.
I am not interested in
Fire-Fighting.
Please do not use
Taguchi Methods for
Fire Fighting
G. Taguchi, 1988

Prevention
Prevention of
Re-occurrence
Fire Fighting

Dr. Genichi Taguchi

2007 ASI Consulting Group, LLC

10

Traditional Product Development

Design-Build-Test Cycle
Vision
Mission
Strategies
&
VOC

Identify
Requirements
&
Establish
Targets

Generate Concepts & Select Concept(s)

Build/Simulate a prototype
Test it to see if it meets
Requirements
Ex. Worst Case Test
Life Test, Etc.

Does it meet Requirements?

Run out of Time/Budget?
Typical Company
spends more than 70% of
Engineering Resource to fire
fight!!
2007 ASI Consulting Group, LLC

Study Design.
Change Design.

No

Yes
Release Design

Manufacturing

Customer
11

Whack-A Mole Engineering!!

2007 ASI Consulting Group, LLC

12

Fire-fighting Proactive Prevention

2007 ASI Consulting Group, LLC

13

CASE STUDY
Robust Optimization with Energy Thinking

1993 Fuel Pump by Ford

System
V = DC Volts

EEC

I = Current

ECT
RPM
Throttle Position
: :
: :
Back Pressure

Scope

Pump

y = Fuel Flow

Injector

P = Back Pressure

Excess hot fuel back to tank

(This is no longer allowed due to pollution standards)

Gas tank
2007 ASI Consulting Group, LLC

14

1993
Fuel Pump by Ford

Ideal Function

I=Current
V=Voltage
P=Pressure

IV
Input Signal:
M
P
Output Response: y Fuel Flow

Ideal Function
y=M

Reality
Very complex

Optimization

After
Optimization

Before
Optimization
M

2007 ASI Consulting Group, LLC

Learning through 1990s

M = IV/P

y = Fuel Flow

Fuel
Pump

Noise Factors

Fuel Temperature, Fuel Type

Tank Pressure, Pump Wear
Driving Conditions, Mfg. Variation, Etc

Reality
Very complex

Ideal Function
y=M
y

2007 ASI Consulting Group, LLC

Response must be a
physical response that
represents Work
generated by the
system
Example: Torque,
Force, Displacement,
Voltage, Current, Flow,
Pressure, Velocity,
Time, 1/Time,
Temp, Pressure,
RPM, Volume
Removed, Work done,
Power, Power
integrated over time,
etc.

16

Learning through 1990s

P-diagram
Control Factors

Modulation Freq., Assembly Type

Motor Design, Valve Design
Mounting Angle, Etc

M = IV/P

Fuel Pump

y = Fuel
Flow

Change
Speed

Noise Factors

Fuel Temperature, Fuel Type

Tank Pressure, Pump Wear
Driving Conditions, Mfg. Variation, Etc

2007 ASI Consulting Group, LLC

17

Ford Fuel Pump

Factors and Levels

Signal Factors and Levels
Level-1 Level-2 Level-3 Level-4 Level-5
6
8
10
12
14
V: Voltage (Volt)
250
300
350
P: Pressure (KPa) 200

M = IV/P

Fuel
Pump

Noise Factors
Noise Strategy (Compunded Noise)

Volatility
Fuel Temp
Vapor Pressure

2007 ASI Consulting Group, LLC

Level-1:
Level-2:
Low Flow High Flow
High
Low
High
Low
0
10

y = Flow

Control Factors
Control Factors and Levels
Level-1 Level-2 Level-3
Tubine Gerotor
A: Pump Type
FDM
Bracket Jet Pump
B: Assembly Type
0
45
80
C: Mounting Angle
Med
High
D: Rated Pump Flow Low
4
9.6
19.2
E: Modulation Freq.

18

Fuel Pump

Inner Array and Outer Array

Control Factors:

Signal Factors: V, P
Noise Factor: N

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18

A
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2

B
2
1
1
1
2
2
2
3
3
3
1
1
1
2
2
2
3
3
3

C
3
1
2
3
1
2
3
1
2
3
1
2
3
1
2
3
1
2
3

D
4
1
2
3
1
2
3
2
3
1
3
1
2
2
3
1
3
1
2

E
5
1
2
3
2
3
1
1
2
3
3
1
2
3
1
2
2
3
1

Inner
Array

e
6
1
2
3
2
3
1
3
1
2
2
3
1
1
2
3
3
1
2

2007 ASI Consulting Group, LLC

e
7
1
2
3
3
1
2
2
3
1
2
3
1
3
1
2
1
2
3

e Outer
8 Array
1
2
3
3
1
2
3
1
2
1
2
3
2
3
1
2
3
1

Assigned to an inner array!

A, B, C, D, E

Assigned to an outer array!

Outer Array

N1

N2

P1
P2
P3
P4
P1
P2
P3
P4

V1
y1
y6
y11
y16
y21
y26
y31
y36

V2
y2
y7
y12
y17
y22
y27
y32
y37

M1 M2 M3 M4
y1 y2 y3 y4

V3
y3
y8
y13
y18
y23
y28
y33
y38

V4
y4
y9
y14
y19
y24
y29
y34
y39

.
.

V5
y5
y10
y15
y20
y25
y30
y35
y40

M40
y40

For each run of L18, one set of

outer array was conducted.
V: Voltage, P: Pressure and N:
Noise are set accordingly.
For each of 40 cell of V, P and
N, measurement was taken on:
o y = Fuel Flow
o I = Current
o P = Actual Pressure
o V = Actual Voltage
For each cell, the value of M is
calculated by V x I / P.
This will result in 40 pairs of M
and y values for each outer
array. S/N is calculated from
this set of data.
19

Fuel Pump
- Result

This performance represents their

accomplishment after 1.5 years of development.

Baseline Design S/N = 17.1 dB

140
120
Flow

100

N1

80

N2

60
40
20
0
0

0.05

0.1

0.15

0.2

0.25

0.3

IV/P

Optimum Design S/N = 28.2 dB

Flow

160
140
120
100
80
60

N1
N2

40
20
0
0

0.05

0.1

0.15
IV/P

2007 ASI Consulting Group, LLC

0.2

0.25

0.3

At the validation, met

all requirements
20

Concept of Signal-to-Noise Ratio Calculation

The least
square best fit
line through
Zero

Signal-to-Noise Ratio is
given by:

F lo w

160
140
120
100
80
60

N1

N2

40
20
0
0

0.05

0.1

0.15
IV/P

2007 ASI Consulting Group, LLC

0.2

0.25

0.3

2
S / N 10 log 2

Where:
= Slope of the best fit line
2 = Mean Square around the
best fit line (Average of
square of distance from
individual point to the best
fit line.)

21

Formulating Successful Robust Design Projects

Formulating a Robust Design Case Study

Robust Design Formulation
1. Define Scope For Optimization
Planning
Stage!!
Very Critical

2. Identify the Ideal Function

3. Develop Signal & Noise Strategy
4. Select Control Factors & Levels
5. Execute & Collect Data
6. Conduct Data Analysis
7. Predict and Confirm
8. Document and Go To Tolerance Design/ Verify

2007 ASI Consulting Group, LLC

22

Cost & Weight Reduction by optimizing function for robustness

Intercooler by Nissan

1996

System & Background

Control Factors
18
1
2

P-diagram
Control Factors
Design Parameters
from Inter-cooler

M=Turbo
rpm

18 inter-coolers were built according to an inner Array of L18

with The outer array was executed for each of 18 runs, as
shown below.
Outer Array

Heat Exchanging meets the

target but it was too noisy.

Signal M

Optimization for Robustness

Noise Factors
Leaks, Temperature,
Humidity, Aging
Position to Position, etc

Where M is set by:

Turbo rpm, Input Air
Volume and Tube Cross
Section Area
2007 ASI Consulting Group, LLC

Inner
Array

T/V RPM 6
Input Air Flow 6
M: Theoretical Airflow 29
Upper - Max Flow 3.4
Upper - Min Flow 2.7
Lower - Max Flow 15.4
Lower - Min Flow 13.6

8
8
6
8
30 40
5.6 6.7
4.4 y 5.4
=
21.8 24.0
18.3 20.0

8 10 10
10 6
8
46 41 50
9.0 8.5 10.7
7.0 Flow
6.5 8.4
Air
26.3 31.3 36.0
21.8 26.8 30.0

10
10
58
12.1
9.8
41.7
34.4

Results and Benefit

Output y
y =Air Flow

Inter
cooler

Ideal Function
y = M

18

ABCDEFGH

Symptoms
Audible Noise
Vibration
Poor Acceleration
Etc.

Ideal Function
y=M

8.8 dB Gain in S/N

and 10% increase
in b was
confirmed!!

Optimum

6 dBA quiter
20% increase in
Heat Exchanging
Efficiency!!
6 dBA

20% more efficient

than requirement,
Scale down to
reduce weight!!
$3.75 Cost Reduction

Baseline

20% increase

2005 Nissan
Steer by Wire

Robust Optimization of New Technology

Standardized S/N for non-linear ideal function & 2-Step Optimization

Noise Factors
Steering Gear

Variation in Level
of Control Factors

Signal

Response

Steering
Feel

Steering Rate

actuator

sensor

Steering Torque

Steering Column
sensor
ECU

Control Factors
System
Components
Devices
Sensors
:

Ideal
Profile
Ideal vs. Baseline Design

7
6
5
4

actuator

Ideal Profile
Baseline N1
Baseline N2

2
1
0
0

50

100

2007 ASI Consulting Group, LLC

150

Ideal Profile

Step-1 OPT N1

Step-1 OPT N2

50

100

150

Ideal Profile

2
0

ECU

After Step-2: Adjust to ideal profile

After Step-1: Reduce Variability

sensors

motor

Step-2 OPT N1
Step-2 OPT N2

0
0

50

100

150

2006 Position Sensor by Alps Electric

Magnetism

Simulation Engine

Scope Big!!

Angle of
incidence

Magnetic Sensing System

M = True
Angle
(0 to 360 Deg)

Sensor

IC

Y = Computed
Angle

Software

S/N Improvement over iterations

L108
48
Control
Factors

49
Noise
Factors

L108

0
-10

x
108x108x360
= 4,199,090
computations

11
iterations

-20
-30

-40

38.7 dB Improvement
Baseline

Optimum

Number of Designs Explored

877,430,873,845,598,000,000,000
True Value (deg)
2007 ASI Consulting Group, LLC

True Value (deg)

2008
Chrysler Frontal Crash

D
e
f
o
r
m
a
b
l
e
B
a
r
r
i
e
r

Front Sill +
Cowl Side
Lower

Shot gun front

Crush Can
Extension

Cradle
Front Leg

Use of 10-min. Spring Mass Model Simulation

vs. 36 Hour Full CAE Simulation

Mid Sill

Rear Sill

Front Tire

Crush
Can

Rail Tip

Cradle Frt X
-member Leg

Shock Tower

Rail Tip Extension

Engine to Front
X-member

Cradle
Middle Leg

OM651
Engine

Rail
Kick down

Main Cradle

Frt Tunnel

A-pillar + Cowl
Side Upper

Mid Rail

Gooseneck
+ Mount

Mid Tunnel

B-Pillar +
Aperture

Rear Rail

Underbody
Stay Brkt Reinf

Rear Tunnel

Loading Zone Mass Legends

Dynamic Operating Window Ideal Function

Dynamic Operating Window Ideal Function
After Optimization

Area
2 you dont
2 want
to deform
Input:
M = Velocity

Output:
Y=Deformation

Output:
Y=Deformation

200

Area you want

1 to deform

Area you want

1 to deform

Dash Intrusions (mm)

180

Intrusion

Before Optimization

W/O Cradle Base

W/O Cradle OPT #1
Cradle Baseline
Cradle OPT #1
Cradle OPT #2
Cradle OPT #3

160
140

120
100

Area
2 you dont
want to deform

80
60
40
20
0

Input:
M = Velocity

Cradle Concept L54 iterated 5 times

K-member Concept L54 iterated 5 times
Short Front Concept L54 iterated 5 times

D-Line Dash C-Line Dash A-Line Dash E-Line Dash

40 N1
40 N1
40 N1
40 N1

D-Line Dash C-Line Dash A-Line Dash E-Line Dash

40 N2
40 N2
40 N2
40 N2

Meeting
European 5-Star
Requirement

# of Designs Explored = 323 x 3 x 5 iterations > 1,400,000,000,000

2007 ASI Consulting Group, LLC

 2007 ASI Consulting Group, LLC

27

It is most important is to recognize

the difference between
Assessment and Validation.
Conduct Assessment first, then Validate
Short Time,
~ One Day

To evaluate how robust

the function is against various
customers usage conditions
Ideal Function + Noise Strategy

Signal-to-Noise Ratio
2007 ASI Consulting Group, LLC

Longer Time,
~ 3 Months.

To check if the product

meets all requirements.
Meet Requirement?
Anything Overlooked?
Any way to add Value?

28

Robust Assessment
M = IV/P

y = Fuel Flow

Fuel
Pump

Noise Factors

Fuel Temperature, Fuel Type

Tank Pressure, Pump Wear
Driving Conditions, Mfg. Variation, Etc

Reality
Very complex

Ideal Function
y=M
y

2007 ASI Consulting Group, LLC

Response must be a
physical response that
represents Work
generated by the
system
Example: Torque,
Force, Displacement,
Voltage, Current, Flow,
Pressure, Velocity,
Time, 1/Time,
Temp, Pressure,
RPM, Volume
Removed, Work done,
Power, Power
integrated over time,
etc.

29

Ideal Function :

Example of Robust Assessment

M=Pressure, y=Flow
Vehicle-A
High Warranty

100.0
80.0
60.0
40.0
20.0
0.0
-0.5-20.0 0
-40.0
-60.0
-80.0
-100.0

O u tp u t F lo w

No
0.0
-2

-1

N1
0

N2

-50.0

O u tp u t F lo w

100.0

-3

Best in
Industry

Best-in-Class

A. High Waranty
50.0

S/N: 3.3 dB
: 30.4

y = M

-2.5

-2

-1.5

-1

-100.0

No

0.5

1.5

2.5

N1
N2

S/N: 12.3 dB
: 36.3

Input Pressure

Input Pressure

Vehicle-B
Lower Warranty

B. Lower Warranty
80.0
60.0
40.0
No

20.0
0.0
-3

-2

-1

-20.0 0

N1
1

N2

O u tp u t F lo w

Output Flow

100.0

S/N: 7.1 dB
: 29.1

-3

-2

-1

-40.0
-60.0
-80.0
-100.0

-3

-2

-1

Input Pressure

No
N1
1

N2

S/N: 6.7 dB
: 31.9

No
N1
1

Competitor
F

Competitor H

N2

O u tp u t F lo w

F lo w O u tp u t

S/N: 8.6 dB
: 32.9

C. Low Warranty
100.0
80.0
60.0
40.0
20.0
0.0
-20.0 0
-40.0
-60.0
-80.0
-100.0

100.0
80.0
60.0
40.0
20.0
0.0
-20.0 0
-40.0
-60.0
-80.0
-100.0
Input Pressure

Input Pressure

Vehicle-C
Very Low Warranty

Competitor
G

Competitor J

-3

-2

-1

100.0
80.0
60.0
40.0
20.0
0.0
-20.0 0
-40.0
-60.0
-80.0
-100.0

No
N1
1

N2

S/N: 10.8 dB
: 12.6

Input Pressure

Make sure our new design is as good as Vehicle-C

2007 ASI Consulting Group, LLC

30

 2007 ASI Consulting Group, LLC

31

Apply Decoupled Development

Step-1: Robust Optimization of Key Technology/ Product
Step-2: Application to family and future products
Step-2
Product
A

New
Technology
New Design
Std Design

Step-1
Robust
Optimization

Step-2
Product
B

Step-2

Robust
Design
Shelf

Product
C
Product
D

Strategies:
Identify Ideal Function based on
Energy Thinking and develop Noise
Strategy
Optimize Ideal Function for
Robustness against Noise Strategy.
2007 ASI Consulting Group, LLC

Product
E

Step-2
Step-2
32

VOC
Strategic
Target
Product
Planning

RE Projects

Validation

Product Development Process

Launch

Step-2

RE Projects
Decoupled Development
- New Technology
- New Design Concept
- Common Components
- Robust BOM/BOD/BOP
- Reuse
- Best Practice
- Design Guideline
- Manufacturing Technology
Robust Assessment
& Optimization
2007 ASI Consulting Group, LLC

Robust
Design
Shelf
(Scalable,
Adjustable &
Robust Design)

Step-1

Note:
Ideally, A should be 20%
and 80% for B.

Only way to reduce Timeto-Market drastically, say

from 48 months down to 12
months!!
Satisfy the VOC in 12
months is easier.
33

(b)
Meet Requirement
by DBTF Cycle

(a)

LSL
LSL

USL

Specification
(Requirement)

USL

Stop!!

y
Enjoy
Cost
(d)
Reduction !

(c)
Robust
Optimization
LSL

USL

LSL

USL

Note: When engineers think their job is just to meet the requirements by Design-BuildTest-Fix Cycle, you are missing numerous opportunities for cost reduction.

2007 ASI Consulting Group, LLC

34

(a)

After
Optimization

LSL

USL

(b)

LSL

USL

Specification
(Requirement)

On the other hand, after optimizing Energy

Transformation for robustness, the design may not meet
Requirements. This an indication of Poor Design
Concept. It is extremely important for engineering
management to detect poor design concept in early stage
of development.
2007 ASI Consulting Group, LLC

35

Thank you for your kind attention!!

Shin Taguchi

2007 ASI Consulting Group, LLC

36