You are on page 1of 18

2

0
0
7

T
h
e

M
a
t
h
W
o
r
k
s
,

I
n
c
.

Design for Six Sigma with MATLAB

Kevin Cohan
The MathWorks, Inc.
2

Design for Six Sigma
http://www.dtic.mil/ndia/2003test/kiemele.pdf
1
10
100
1000
R
e
l
a
t
i
v
e

C
o
s
t

t
o

M
a
k
e

a
D
e
s
i
g
n

C
h
a
n
g
e
Product Stage
Research Design Development Production
Classic Six Sigma
focuses here
DFSS focuses here
3

DMAIC Methodology
Define the problem / defects
Measure the current performance level
Analyze to determine the root causes of the problem / defects
Improve by identifying and implementing solutions
that eliminate the root causes
Control by monitoring the performance of the improved process
M
D
A
I
C
http://www.dtic.mil/ndia/2003test/kiemele.pdf
D M A I C

2
0
0
7

T
h
e

M
a
t
h
W
o
r
k
s
,

I
n
c
.

Example
Optimization of an
Engine Cooling Fan Design
Image courtesy of Novak Conversions
5

Define:
Define the Problem
Existing engine isnt cooled sufficiently
under difficult driving conditions
New design requirement
Airflow > 875 cubic feet per minute
Approach
Optimize design factors
for maximum airflow
Use the MATLAB product family to
implement the DMAIC methodology
D M A I C
Image courtesy of Novak Conversions
6

Measure:
Measure Cooling Fan Performance
Import historical data
from Excel
Establish baseline
for comparison
Mean = 842 ft
3
/ min
Std Dev = +2 ft
3
/ min
D M A I C
7

Analyze:
Analyze Factors that Affect Performance
Three design factors
we can modify
Clearance
Distance
Pitch
Min Max
Distance from radiator (d) 1.0 1.5 inches
Blade pitch angle (p) 15 35 degrees
Blade tip clearance (c) 1.0 2.0 inches
Factors
Range
Units
D M A I C
8

Analyze:
Analyze Factors that Affect Performance
Gather additional data from existing system
Modify our design factors and measure performance
Fit a model to the performance data
Evaluate the model to understand the
effect of each design factor
What tests do we run?
and how many?
D M A I C
(
d
)
(
p
)
(
c
)
Test Run
D
i
s
t
a
n
c
e

f
r
o
m
r
a
d
i
a
t
o
r
B
l
a
d
e

p
i
t
c
h
a
n
g
l
e
B
l
a
d
e

t
i
p
c
l
e
a
r
a
n
c
e
P
e
r
f
o
r
m
a
n
c
e
1 ? ? ?
2
3

n
9

Analyze:
Analyze Factors that Affect Performance
Use a designed experiment to gather performance data
Choose the Box-Behnken function to determine
points in the design space
D M A I C
10

Analyze:
Analyze Factors that Affect Performance
Fit a quadratic model to the data (regstats function)
D M A I C
AF = B
0
+ B
1
X
1
+ B
2
X
2
+ B
3
X
3
+
B
4
X
1
X
2
+ B
5
X
1
X
3
+ B
6
X
2
X
3
+
B
7
X
1
2
+ B
8
X
2
2
+ B
9
X
3
2
AF = Airflow (ft
3
/min)
X
1
= Distance from radiator (inches)
X
2
= Fan pitch angle (degrees)
X
3
= Tip clearance between fan blades
and shroud (inches)
11

Analyze:
Analyze Factors that Affect Performance
Use rstool to visually inspect interaction between all three design
factors (distance, pitch, clearance) simultaneously
D M A I C
12

Improve:
Improve the Cooling Fan Performance
Use optimization to automate the task of
finding the maximum airflow
fmincon function
D M A I C
13

Improve:
Improve the Cooling Fan Performance
Use uncertainty analysis to ensure a robust design
Two main contributors
Model uncertainty
Manufacturing variability
D M A I C
Model noise
0.00 +/- 0.96 ft
3
/min
Distance from radiator (d) 1.00 +/- 0.005 inch
Blade pitch angle (p) 27.3 +/- 0.5 degrees
Blade tip clearance (c) 1.00 +/- 0.005 inch
Factors Nominal Value and Tolerance
14

Improve:
Improve the Cooling Fan Performance
Use Monte Carlo simulation to determine
impact of these variations
Mean = 882 ft
3
/ min
Std Dev = +2 ft
3
/ min
D M A I C
15

Control:
Control the Manufacturing and Installation
Use Statistical Process Control (SPC)
techniques to monitor and
control manufacturing and
installation of the fan
D M A I C
16

Summary
Continued emphasis on quality initiatives
Data analysis increasingly becoming an
integral part of the engineering design process
MATLAB product family can
support these initiatives

2
0
0
7

T
h
e

M
a
t
h
W
o
r
k
s
,

I
n
c
.

Questions?

2
0
0
7

T
h
e

M
a
t
h
W
o
r
k
s
,

I
n
c
.

You might also like