Course Introduction
Probability, Statistics and
Quality Loss
HW#1 Presentations
Robust System Design
16.881
Session #1
MIT
Background
? SDM split-summer format
– Heavily front-loaded
– Systems perspective
– Concern with scaling
? Product development track, CIPD
– Place in wider context of product realization
? Joint 16 (Aero/Astro) and 2 (Mech E)
Robust System Design
16.881
Session #1
MIT
Course Learning Objectives
? Formulate measures of performance
? Synthesize and select design concepts
? Identify noise factors
? Estimate the robustness
? Reduce the effects of noise
? Select rational tolerances
? Understand the context of RD in the end-to-end business
process of product realization.
Robust System Design
16.881
Session #1
MIT
Instructors
? Dan Frey, Aero/Astro
? Don Clausing, Xerox Fellow
?Joe Saleh, TA
? Skip Crevelling, Guest from RIT
? Dave Miller, Guest from MIT Aero/Astro
?Others ...
Robust System Design
16.881
Session #1
MIT
Primary Text
Phadke, Madhav S., Quality Engineering
Using Robust Design. Prentice Hall, 1989.
Robust System Design
16.881
Session #1
MIT
Computer Hardware & Software
? Required
– Access to a PC running Windows 95 or NT
– Office 95 or later
– Reasonable proficiency with Excel
? Provided
– MathCad 7 Professional (for duration of course
only)
Robust System Design
16.881
Session #1
MIT
Learning Approach
? Constructivism (Jean Piaget)
– Knowledge is not simply transmitted
– Knowledge is actively constructed in the mind
of the learner (critical thought is required)
? Constructionism (Seymour Papert)
– People learn with particular effectiveness when
they are engaged in building things, writing
software, etc.
– http://el.www.media.mit.edu/groups/el/
Robust System Design
16.881
Session #1
MIT
Format of a Typical Session
?
Lecture
Well, almost
? Reading assignment
?Quiz
? Labs, case discussions, design projects
? Homework
Robust System Design
16.881
Session #1
MIT
Grading
? Breakdown
– 40% Term project
– 30% Final exam
– 20% Homework (~15 assignments)
– 10% Quizzes (~15 quizzes)
? No curve anticipated
Robust System Design
16.881
Session #1
MIT
Grading Standards
Grade Range Letter Equivalent Meaning
97-100 A+ Exceptionally good performance demonstrating
superior understanding of the subject matter.
94-96 A
90-93 A-
87-90 B+ Good performance demonstrating capacity to use
appropriate concepts, a good understanding of
the subject matter and ability to handle
problems.
84-86 B
80-83 B-
77-80 C+ Adequate performance demonstrating an
adequate understanding of the subject matter, an
ability to handle relatively simple problems, and
adequate preparation.
74-76 C
70-73 C-
67-70 D+ Minimally acceptable performance
demonstrating at least partial familiarity with the
subject matter and some capacity to deal with
relatively simple problems.
64-66 D
60-63 D-
<60 F Unacceptable performance.
Robust System Design
16.881
Session #1
MIT
Reading Assignment
? Taguchi and Clausing, “Robust Quality”
?Major Points
– Quality loss functions (Lecture 1)
– Overall context of RD (Lecture 2)
– Orthogonal array based experiments (Lecture 3)
– Two-step optimization for robustness
? Questions?
Robust System Design
16.881
Session #1
MIT
Learning Objectives
? Review some fundamentals of probability
and statistics
? Introduce the quality loss function
? Tie the two together
? Discuss in the context of engineering
problems
Robust System Design
16.881
Session #1
MIT
Probability Definitions
? Sample space - List all possible outcomes of an
experiment
– Finest grained
– Mutually exclusive
– Collectively exhaustive
? Event - A collection of points or areas in the
sample space
Robust System Design
16.881
Session #1
MIT
Probability Measure
? Axioms
– For any event A, P( A) ≥ 0
– P(U)=1
– If AB=φ, then P(A+B)=P(A)+P(B)
Robust System Design
16.881
Session #1
MIT
Discrete Random Variables
? A random variable that can assume any of a
set of discrete values
? Probability mass function
– p
x
(x
o
) = probability that the random variable x
will take the value x
o
– Let’s build a pmf for one of the examples
? Event probabilities computed by summation
Robust System Design
16.881
Session #1
MIT
Continuous Random Variables
? Can take values anywhere within
continuous ranges
? Probability density function
b
–
P{a < x ≤ b}=
∫
f
x
( x)dx
a
–
0 ≤ f
x
( x) for all x
∞
–
∫
f
x
( x)dx = 1
?∞
Robust System Design
16.881
Session #1
MIT
Histograms
? A graph of continuous data
? Approximates a pdf in the limit of large n
Histogram of Crankpin Diameters
5
Frequency
0
Diameter, Pin #1
Robust System Design
16.881
Session #1
MIT
Measures of Central Tendency
b
? Expected value E( g( x)) =
∫
g( x) f
x
( x)dx
a
? Mean μ = E(x)
n
? Arithmetic average
1
∑
x
i
n
i=1
Robust System Design
16.881
Session #1
MIT
Measures of Dispersion
?Variance
VAR( x) =σ
2
= E(( x ? E( x))
2
)
? Standard deviation
σ =
n
) ))( ((
2
xEx E ?
? Sample variance
S
2
=
1
∑
( x
i
? x )
2
n ? 1
i?1
? n
th
central moment
E(( x ? E( x))
n
)
? n
th
moment about m E(( x ? m)
n
)
Robust System Design
16.881
Session #1
MIT
Sums of Random Variables
? Average of the sum is the sum of the
average (regardless of distribution and
independence)
E( x + y) = E( x) + E( y)
? Variance also sums iff independent
σ
2
( x + y) =σ( x)
2
+σ( y)
2
? This is the origin of the RSS rule
– Beware of the independence restriction!
Robust System Design
16.881
Session #1
MIT
Central Limit Theorem
The mean of a sequence of n iid random
variables with
– Finite μ
–
E ( x
i
? E( x
i
)
2+δ
) < ∞ δ > 0
approximates a normal distribution in the
limit of a large n.
Robust System Design
16.881
Session #1
MIT
Normal Distribution
2
1
( x )
e( x)
σ π
?
?μ
σ
2
f
x
2
=
2
μ
99.7%
68.3%
1-2ppb
Robust System Design
16.881
Session #1
MIT
Engineering Tolerances
? Tolerance --The total amount by which a
specified dimension is permitted to vary
(ANSI Y14.5M)
? Every component
p(y)
within spec adds
to the yield (Y)
L
U
Y
y
Robust System Design
16.881
Session #1
MIT
Crankshafts
? What does a crankshaft do?
? How would you define the tolerances?
? How does variation affect performance?
Robust System Design
16.881
Session #1
MIT
GD&T Symbols
Robust System Design
16.881
Session #1
MIT
Loss Function Concept
? Quantify the economic consequences of
performance degradation due to variation
L(y)
What should the function be?
y
Robust System Design
16.881
Session #1
MIT
Fraction Defective Fallacy
? ANSI seems to imply
a “goalpost”
L(y)
mentality
? But, what is the
difference between
A
o
– 1 and 2?
– 2 and 3?
3
2 1
Isn’t a continuous function
m-?
ο
m
m+?
ο
y
more appropriate?
Robust System Design
16.881
Session #1
MIT
A Generic Loss Function
? Desired properties
– Zero at nominal value
– Equal to cost at
specification limit
– C1 continuous
? Taylor series
∞
f ( x) ≈
∑
1
( x ? a)
n
f
( n )
(a
L(y)
A
o
m-?
ο
m
m+?
ο y
n=0
n!
)
Robust System Design
16.881
Session #1
MIT
? Defined as
L(y)
Nominal-the-best
L( y) =
A
o
2
( y ? m)
2
?
o
A
o
? Average loss is
proportional to
the 2
nd
moment
m-?
ο
m
m+?
ο
y
about m
quadratic quality loss function
(HW#2 prob. 1)
"goal post" loss function
Robust System Design
16.881
Session #1
MIT
Average Quality Loss
L(y)
m
m+?
[L E
σ
μ
A
o
( y)] =
A
o
2
[σ
2
+ (μ? m)
2
]
?
o
m-?
ο
ο
y
quadratic quality loss function
probability density function
Robust System Design
MIT16.881
Session #1
Other Loss Functions
? Smaller the better L( y) =
A
o
2
y
2
?
o
(HW#3a)
? Larger-the better
L( y) = A
o
?
o
2
1
2
y
(HW#3b)
? Asymmetric
A
o
2
( y ? m)
2
if y > m
(HW#2fc)
L( y) =
?
Upper
A
o
2
( y ? m)
2
if y ≤ m
?
Lower
Robust System Design
16.881
Session #1
MIT
Printed Wiring Boards
? What does the second level connection do?
? How would you define the tolerances?
? How does variation affect performance?
Robust System Design
16.881
Session #1
MIT
Next Steps
? Load Mathcad (if you wish)
? Optional Mathcad tutoring session
– 1hour Session
? Complete Homework #2
? Read Phadke ch. 1 & 2 and session #2 notes
? Next lecture
– Don Clausing, Context of RD in PD
Robust System Design
16.881
Session #1
MIT