I. Overall Strategy
II. Making Quantum Improvement with
Simple DOE
III. Going After Basic Robustness
IV. Leaving No Stones Unturned Building
Robustness for Expanded Application
Capabilities
V. Qualitek-4 software for design and
analysis tasks (time permitting)
Ref. Page 1-1
Ref: The Evolution of Six Sigma, by Dr. S. Marash, Quality Digest, June
2001
Note: For some reason the author did not include ISO/QS 9000 which was popular
quality disciplines implemented by most manufacturing companies.
Ref. Page 1-2
DOE
SPC
FMEA Taguchi
Approach
Ref. Page 1-2
What is DOE
Return on
Investment
* Production
* Test &
Validation
* Design &
Development
Where do we do quality
improvement?
* Design &
Analysis
Driving Questions For Quality Improvement
(Opportunities for Building Quality)
* Production
* Test &
Validation
* Design &
Development Is the performance at its best or at optimum?
* Production
* Test &
Validation
* Design & Will the product perform under
Development extremes of application environment?
Yavg.
Yo
Ref. Page 1-8
PLAN
Define project
Determine performance
Need to follow a
objectives
structured
approach Identify factors to study
CONFIRM ACHIEVEMENTS
No new knowledge Test predicted design and verify that the
required
performance achieved is acceptable.
Ref. Page 1-10
EXAMPLE APPLICATION
It is an experimental technique that determines the
solution with minimum effort.
In a POUND CAKE baking process with 5 ingredients,
and with options to take HIGH and LOW values of
each, it can determine the recipe with only 8
experiments.
Full factorial calls for 32 experiments. Taguchi
approach requires only 8
Ref. Page 1-11
Factors Level-1
Level-2
A1 A2
A: Egg
B: Butter B1 B2
C: Milk C1 C2
D: Flour E1 E2
E: Sugar D2 D1
FIVE factors at TWO levels each make 25 = 32 separate recipes (experimental condition)
of the cake.
Ref. Page N/A
Experimental Conditions
A1 B1 C1
Condition #1 D1 E1
A1 B1 C1
Condition #2 D1 E2
Ref. Page N/A
Experimental Conditions
A1 B1 C1
Condition #3 D2 E1
A1 B1 C1
Condition #4 D2 E2
Ref. Page N/A
Experimental Conditions
A1 B1 C2
Condition #5 D1 E1
A1 B1 C2
Condition #6 D1 E2
Ref. Page N/A
Experimental Conditions
A1 B1 C2
Condition #7 D2 E1
A1 B1 C2
Condition #8 D2 E2
Experimental Conditions
Condition # 9 through 30 . . . . .
Ref. Page N/A
Experimental Conditions
A2 B2 C2
Condition #31 D2 E1
A2 B2 C2
Condition #32 D2 E2
Ref. Page N/A
Experimental Trial Conditions by L-8 Orthogonal
Array
Ref. Page N/A
Fishing Net
Ref. Page 1-12
II. Experiment
Stamping / Design & ResultsDeburring Rust
Hobbing Clutch plates Inhibitor
One 4-level factor
Clutch plate and four 2-levelare
factors
tumbled in this experiment
Parts were
are studied using a
modified L-8
madearray.
from The 4-level factor submerged
was assigned to column 1 modified using original
in a large
1/16 inch container to in a
columns 1,thick
2, and 3.
rolled remove sharp chemical
steel edges bath
Cleaned and dried parts
are boxed for shipping.
Applications in Analytical Simulations
F
d L
Elasticity, EI
W
Ln
5 2 1 2 1 2 1
(XY)
2 1 2 2
2 2
Use this array (L- 3 1 3 3 6 2 1 2 2 1 2
9) to design 3 1
experiments with 4 2 1 2 7 2 2 1 1 of 2columns
No. 2
1 in the array.
four 3-level 3
5 2 2 3 8 2 2 1 2 1 1
factors
1 2
No. of rows No. of levels
6 2 3 1
in the array in the
2
columns.
7 3 1 3
2
8 3 2 1
3
9 3 3 2
There are More Steps to Climb
Interactions
Mixed level factors
Result/Response
Result/Response/
/QC
QC
A1 A2 A3
A1 A2 A4
A3
Ref. Page 2-
Combination Possibilities Full Factorial 3
Combinations
NOTATIONS:
A (A1,A2) or A represent 2-level factor
Cond.# A B C
1 1 1 1
THREE 2-level
ONE 2-level factors
factor offer create
TWO test conditions (A ,A ).
1 2 2 1 1 2
TWO 2-level factors create FOUR (22 = 4 )
EIGHT (2 = 8) possibilities.
3
test conditions A1B1 A1B2 A2B1 A2B2) .
3 1 2 1
A1B1C1 A1B1C2 4 1 2 2
A1B2C1 A1B2C2 Simpler 5 2 1 1
notations for all
A2B1C1 A2B1C2 possibilities or 6 2 1 2
full factorial
A2B2C1 A2B2C2
7 2 2 1
8 2 2 2
Ref. Page 2-
Full Factorial Experiments Based on Factors and 4
Levels
3 Factors at 2 level 23 = 8
4 Factors at 2 level 24 = 16
7 Factors at 2 level 27 = 128
15 Factors at 2 level 215 = 32,768
2-Level Arrays
L-4 Orthogonal Array
L4 (23 ) Trial # 1 2
L8 (27) 3
L12 (211) 1 1 1 1
L16 (215) . . . . 2 1 2 2
3 2 1 2
4 2 2 1
3-Level Arrays
L9 (34), L18 (21 37) . . .
4-Level Arrays
L16 (45) . . . .
Ref. Page 2-
5
Properties of Orthogonal Arrays
Key observations:
First row has all 1's. There is no row that has all 2's.
All columns are balanced and maintains an order.
Columns of the array are ORTHOGONAL or balanced. This means that there are
equal number of levels in a column. The columns are also balanced between any
two columns of the array which means that the level combinations exist in equal
number.
Within column 1, there are two 1's and two 2's.
Between column 1 and 2, there is one each of 1 1, 1 2, 2 1 and 2 2
combinations.
Factors A, B And C all at 2-level produces 8 possible combinations (full factorial)
Taguchis Orthogonal array selects 4 out of the 8.
How does One-Factor-at-a-time experiment differ from the one designed using an
Orthogonal array?
Ref. Page 2-
Orthogonal Arrays for Common Experiment 6
Designs
3 2 1 2 xxx
4 2 2 1 xxx
Orthogonal Arrays for Common Experiment Ref. Page 2-
6
Designs
Trial#
1
2
A
1
1
B
1
1
C
1
1
D
1
2
E
1
2
F
1
2
G
1
2
Result
s
xx
xx
Ln (XY)
No. of columns
in the array.
3 1 2 2 1 1 2 2 xx
No. of rows in No. of levels in
4 1 2 2 2 2 1 1 xx the array the columns.
5 2 1 2 1 2 1 2 xx
6 2 1 2 2 1 2 1 xx
7 2 2 1 1 2 2 1 xx
8 2 2 1 2 1 1 2 xx
Orthogonal Arrays for Common Experiment Ref. Page 2-
7
Designs
Ln (X Y)
3 1 3 3 3 xx
4 2 1 2 3 xx
5 2 2 3 1 xx
6 2 3 1 2 xx
7 3 1 3 2 xx No. of rows in No. of levels in
the array the columns.
8 3 2 1 3 xx
9 3 3 2 1 xx
Ref. Page 2-
7
Steps in Experiment Design
L8 Orthogonal Array
Control
Factors
Trial# A B C D E F G
1 1 1 1 1 1 1 1
2 1 1 1 2 2 2 2
3 1 2 2 1 1 2 2
4 1 2 2 2 2 1 1
5 2 1 2 1 2 1 2
6 2 1 2 2 1 2 1
7 2 2 1 1 2 2 1
8 2 2 1 2 1 1 2
Inner
Array
Ref. Page 2-
Full Factorial Arrangement with Seven 2-level Factors 9
Ref. Page 2-
10
Common Orthogonal Arrays
2-Level Arrays
L4 (23)
L8 (27)
Ln (XY)
No. of columns
in the array.
L12 (211)
L16 (215) No. of rows in No. of levels
the array in the columns.
3-Level Arrays
L9 (34)
L18 (21 37)
4-Level Arrays
L16 (45)
Ref. Page 2-
10
Planning Before Designing Experiments
PLAN
Identify Project and Select Project Team
Define Project objectives Evaluation Criteria
CONDUCT EXPERIMENTS
ANALYZE RESULTS
Factor Effects, Optimum Condition, Predicted
Performance, etc.
Ref. Page 2-
11
Trial# C A B
Results
1 1 1 1 5
2 1 2 2 8 __
T = (5 + 8 + 7 +4)/4 =
3 2 1 2 7
6
4 2 2 1 4
Trial# C A B _
Results
1 1 1 1 5 _A1 = (5 + 7)/2 = 6.0
2 1 2 2 8
3 2 1 2 7 _
_A2 = (8 + 4)/2 = 6.0
4 2 2 1 4
Ref. Page 2-
14
Analysis of Experimental Results
Quality Characteristics
Examples
Nominal is Best: 5 dia. Shaft,12 volt battery, etc.
Smaller is Better: noise, loss, rejects, surface roughness, etc.
Bigger is Better: strength, efficiency, S/N ratio, Income, etc.
Estimate of Performance at the Optimum Ref. Page 2-
15
Condition
U 9 Main Effects
NP (Also called: Factorial Effects or Column Effects)
OP 8
PE
D
7
6
KE
R
A1 B1
5 C2
N
EL 4
S
3
A1 Hot plate A2 B1 Oil B2 C1 Heat Setting
C2
__ __ __ __ __ _ __
Yopt = T + ( A1 - T ) ( B1 - T ) ( C2 - T )
= 6.0 + ( 6 +6 ) + (4.5 + 6.0 ) + ( 5.5 6.0
)
= 4.0 (Assumption: Factor contributions are additive)
Interpretation of the Estimated Ref. Page 2-
16
Performance
Expected Performance:
What is the improved performance?
How can we verify it?
What is the boundary of expected performance?
(Confidence Interval, C.I.)
Notes:
Generally, the optimum condition will not be one that has already been tested. Thus you will need to run additional
experiments to confirm the predicted performance.
Confidence Interval (C.I.) on the expected performance can be calculated from ANOVA calculation. These boundary
values are used to confirm the performance.
Meaning: When a set of samples are tested at the optimum condition, the mean of the tested samples is expected to be
close to the estimated performance.
(Yopt - YCurrent
Improvement = x
)
100 YCurrent
(4 - 6 )
= x 100 = - 33%
6
Practice and Learn
Solve
Problem 2A
Ref. Page 2-41
Practice Problem # 2A
[Answers: 1 - (Describe, Factor __ has the most influence, etc. ) 2 - Optimum Cond: 2,2,2, , 3 - Gd. Avg.=28.5, 4-
Yopt = 19, 5 - Contribution = 9.5, 6 - Improvement = 24% ]
Group Exercise - Class Project
I. Experiment Planning
Project Title -
Objective & Result -
(Describe why you initiated the project and what you wish to accomplish)
Quality Characteristics: (Describe what you are after and how you would measure the
results. Depending on what it is you are after, your quality characteristic will be bigger
is better, smaller is better, or nominal is the best)
Noise Factors
Outer Array
Humidity 31 2 2 1 L4
Temperature 2 1 2 1 2
Oven type 1 1 1 2 2
Control factors
Tr# 1 2 3 4 5 6 7 1 2 3 4
Results
1 R 11 R R R
2 R 21 R R R
3 .......
L8 . . .. . . .. .. . . .. .
*
R R R73 R
8 Inner Array R R R R
Evaluating Performance Based on Reduced Variation
Why -(Minus)
Why base 10
Leaving no Stones Unturned Building Robustness
for Expanded Application Capabilities
Representing the
subject
product/process design
as a system is a
necessary step for
understanding its
behavior. The system is
described in terms of
its parameters like
factors, signal,
response, noise, etc. An
absence of variable
input/signal to the
system renders it as a
static system which are
dealt with the standard
designs discussed
earlier.
Example of Dynamic System Response
y
Example Case: Fuel Gauge Reading
Nonlinear
High
variation *
Response *
M y = *
Least Desirable (y) M *
* *
y *
*
* *
, *
* Slope
*
Linear *
Large variation *
M1 M2 M3
Signal (M)
M
Comparison of Strategies for Static and Dynamic
System
Vijay Singh has been on Tiger's heels for some time now, and if
either Singh, or Ernie Ells were to win the PGA Championship,
Woods would relinquish his spot atop the rankings.
*
Signal Factor (M): Response *
Battery voltage *
(y) *
Response (y): * y = *
Voltage reading * M
*
Notes: * *
* , *
Slope
*
*
*
M1 M2 M3
Signal (M)
Dynamic System Examples
*
Signal Factor (M): Response *
Weight of subject *
(y) *
Response (y): * y = *
Indicated weight * M
*
* *
Notes: * , *
Weight indicated should be the same Slope
*
as weight of the subject. *
*
M1 M2 M3
Signal (M)
Where to Apply and Why
WHY?
Very high return on investment (ROI)
Presented
by
Ranjit K. Roy, Ph.D., P.E. PMP, Fellow of ASQ., President, Nutek, Inc.
And
Suren N. Dwivedi,,Ph.D., P.E., Professor, Dept. of Mechanical Engineering
University of Louisiana at Lafayette