This preview shows pages 1–19. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: Taguchi Methods and ‘ Response Surface ' Alternatives Once in control,
goal is to produce
uniformly near a
target. I ° Mean on target ° Reduce variation Three possible
situations:
 Smaller is better 0 Larger is better ° On target is best On target is best.
Taguchi considered
deviation from target to be loss to
society. Taguohi
models this as: 120’) =K(Y—T)2
Y 2 Quality Control ' T 2 Target value On average, loss is
proportional to: ‘ E(Y — T)2
= Varﬂ’)
+ (Bier?)2
where
B iaS : difference of mean and Target 5 For smaller is better,
Taguehi has less as: L(Y) = Y2
_ i.e. on average
E(Y 2).
So Target is 0. For larger is better,
Taguchi considers
loss as: MY) =1/y2
0f E(1/y2). First step in Taguchi
' Method is parameter % I design. Parameters
are inputs known or
suspected to affect the
quality characteristic. Goal: Design in
quality, don’t sample
to catch poor output. 8 Parameters are divided
into two groups: ° Control variables
parameters which can be
controlled in the process. ° Noise variables—
parameters which vary and
are usually not controlled (sometimes hard or
expensive to control). Levels are chosen
for each of the
control variables. Levels are chosen 1 and temporarily
ﬁxed for each of the  noise variables. Control variables are placed into inner array.
Example: [\JNHMNNHh—d :Nl—tNr—tNr—twv—‘l Eight design points, each
row one design point 11 Noise variables are
' placed in outer array Example: '
2 l
2 2 Generally the levels span
What is experienced in
production conditions. 12 The structure of each
array is typically ° Factorial design ° Fractional factorial design
(sometimes referred to as
 “orthogonal array”.) The arrays are then
crossed and observa—
tions collected. 13 Goal of analysis:
Minimize
_ E(Y — N by choice of best control
variable settings. Taguohi suggests look at
“signal to noise” ratio. % Bigger is better. Then
adjust mean. 14 Criticisms: ° Crossing inner and
outer arrays can lead to
unwieldy designs. ° Use of signal to noise
ratios very adhoc and
hard to generalize
(various versions). '15 Alternatives proposed
include ReSponse
Surface Methodology
' (RSM). * Goals:  Model the response Y as a
function of control and noise
variables.  Use information on distribution
of noise variables to model
mean and variance. l6 Simple example: I Y = Lamina thickness
X I : Viscosity (control variable) _ X 2 :— Ambient humidity (noise variable) . Fitted model:
Y = 60+B‘1X1+B\12X1X2 Select X to minimize loss. 17' E(f)=6O+BIXI ‘
+6 X E(X2) 12 1 Var()7) 2(612X1r Var(X2) 2 +6 (Conditional mean and
variance, conditional 0n
~ parameter estimates). 18 More complex example: XPX2 —C0ntr01 X3 —N0ise 17:50 +61X1+BA2X2 + B23X2 'X3 E®Zﬁo+61X1+62X2+323X2+623X2 '(EX3)
Vam=(ﬁ23X2)Vaz(X3)+o Can set X2 set X1 to control mean. to minimizevariance 19' ...
View
Full
Document
This note was uploaded on 02/20/2012 for the course STAT 513 taught by Professor Na during the Spring '11 term at Purdue.
 Spring '11
 NA

Click to edit the document details