You are on page 1of 6

12-5 THE TAQUCHI METHOD

In this wction we don1 with the philosophy and experimontnl dcsign principles devised
by Geniehi Tnguchi, n Japanewe engineer whosc idem in quality cnginccring llrtvc bcen
wed tor many ycnto in Japan. The underlying theme of the ncvcrsndfng cyclc ofqunl-
ity improvement is wppwted by Taguchik work Wguchi and Wu 19W,Tuguchl 1986.
19871, QtuUty enghcerlng ha# the ohjcctive of deigning qudlly into cwry product
and corresponding process. It directs quality improvement eflorta upstream [ram the
manufacturing praccse to the design phase and is thercforc referred to w an off-linc
@I@ control methd. The techniques of rtatistical prwers eontml discussed in prc-
ceding chapters are known as on-line quality control methods. Taguchib off-line meth-
od# are effective ip Improving quality and cutting down coats at the same time
(Taauetri, Elsayed, and Hriang 1989). Off-line math& improve product manufactura-
bility and reduce p d u c t development and lifetime costs.
Ip Tamchi$ method, quality in measured by the deviation of a characteristic
from Its target value. A Ima hietion is dtvclopad for this deviation. Uncontrollable
factors, known as noise,cause such deviation and thereby lead lo Imr. Since the ciimi-
nation of noiee lacton is impractical nnd often irnp~sible,the 'hguchl mclhwl seeks
to minimize the effects of noise and to determine the optimal level of thc imporiant
wntrolhble factors bawd on the concept of mbastn&m(Dehnad 1989: Nair 199.2). The
objective Is to create a p r d u c t l p r ~ c edesign
~ that is instnsltiw to all possible combi-
nations of tho uncontrolhibIe rioisc factors and is nt the same time effective and cost-
clTkiant as n result of getting the key mntrollahle facton at certain Irvela. Whercns the
fdndations of Taguchlb quality engineering sccm to be well accepted, there nre criti-
cl~mm,which we will diwass later. of the stntisticai aspects of hiti cxpcrimcntal dcsigns.

12-6 THE TAGUCHI PHILOSOYHV


' According lo Taguchi, "Quality ir the loss imparted Lo society From the timt n product
ir shipped."Taguchi attaches a monctary vaIue l o quality bccause he feeh t h ~ this
t will
make quality improvcment understood by all-technical penonnel as we11 as manage-
ment. Qpical cxamplcs of lost to society include Znilurc.to meet the eustomcr's re-
quirements and unsatisfactory performance thet lends to lots d goodwi ll and reduced
- : '1
1

!
market #hare. Thc TBguchl wnwpt of qrtaIity Iwr ~ I U bcan'cxttrided by Kackolr (19861
CHAPTER 12 to include the lm'to rodsty *lie the produet 'libins manufictured For instance,
the raw materials and labor dspletad when maklag tho product or rendering the str-
v$c may rlro b &dared sdataI lawas. Llkewlw, the mcs lr olenrironmenralpollu-
tion fallb into this ategot):
The p u w of quality improvement, then, is to discover Innovati* nays dl dc-
aigning p r o d m and processes that will save s e r y more than thcy cost in the Iang '
run. All prducta ultlrnatcty causa IOU becnue they break and then need to be re
paired Or replaced or h u s e they wear out alter an adequate performan- in their
functional life. The d a m to lphieh the produe or wrvice m6sh copsumerr' cxpcda-
tions alrcct8 the rnfignltude of ION.TBguchl catondr thnt~thcloss duc tct a produak
uarintion in perlomnoe is proportiqnnl fo the qusre of the pformrnce characteris-
tic%deviation troy I~ targPt VIIUC.
Figure 1241 shbwa m,exrmplsof the quadrntic lmfunctlon. ~ h i funkioa'js a h ,

cxampje of tho target-ls-kt'etmdition, IF the quality charactetistic y i~ at the target


vaiue m, rhon the fosst r y ) ir minitnized. As the chiractcriatlc deviates from the 4aqd ..
(oinmmha1) viitw in either d h t h , the 1- incruses in a quadratic mmnor. Exnm- -

plea imetnde the dimension* ofa part (length, thickners. diarnbter, and so am) and the
q c d n g pressurn in a pm8.
The quality and cmt of8 mulaaured p#ducl S infl"end by the cnghetring de-
s i p d tbe product as mll a# the process. By conccntratlng-bn p d w t and proecss de*
,
tip that arc mbua-that is, t&& sensitivt to uncmtrolkble tactom such as kmpmture,
h d * , and man- , d a t S w , ( r h t nohe factors), the Tnguchi concept attempts
. to reduce the impart of nofss ratherjhan eiirnimk it. Frequently, the elimination or no& , -

factors Iu neither p r a W it is wry mtIy) nar ftasibIe. Manufacturing variabiI-


ity amnot totally e l h h t ~ $ha,miq umpwlants and row matetial hhermtIy tx-
hibit varlatiw. By BampurIng thc impet ,d the nokc facton and by doering the '
ccmWd8bk iactar k w b that force the desirable quality chtactcristles to amy dose to
target Wuw. a robust Wgr d the p d w t , ahd,thefebyof the pmxm. is d i e d .
Taguchi advmns a three-stap design operation to determine the target .values
and toleranccr €or relevant paramatem in the product and the procasa: system design,
parameter design. nnd tolerance design, In e rw,
n scientincand cngincering
prinelpha and ~xperieneaare uded to create a prototyp~of tha product that will mast :
the fbnctionai requirements and alga tocreate the process that wilt build it. - '
Pammeter d* lnvdvg Rndiag tht optimal settings of the product and
process parameters in order to minim&k performance variabili* Thguchi defints a ,<

I
1
1
I
-*

5
&
%'
3
m-4 m nr*h
FiGUAE 124 1 Lorr hmctbn for a shmtlon in whloh th* Torpat vbu4
we vdur h b t . . a~dhyc h m n c ~ r lyl
~e
the sour- of variation. This is the kcy to nchieving robulrt dcrigns in the Tnguchi
mcaod. This method b cort~efficientin tlint i t redtrees the prformance variabilily by
mlnimidng the effects of factors wc cannut control instead ol trying to control or elim-
iruto them,
Our example also illustrates the fundamental idea ofthe two-srep procedure used
k the nmeter duign phanc. In t h first ~ step, we choose parameter levels that mini-
$=
mkr e variability of thd output quality charackrtstic. T h i s step makes we of the
n o d w a r offecta of the praa#(m on the response variable atid creates *.design that
b r o k t Hth rsrrpect to the unwntrolInhIe sources of variation. In the sec~nd~step. we
identify paratnetm~that have a h c n r ofIect on the average value of the response but
do ria iRlgrCt variability, Thest pnmrneters, known as adjustment factors, ere used to
drive ths q d h y charactc~ticto the-targut value wi thou?incresaing \he variability.
Aft# tht systemdesign and parnmctcr derign stages comes the third stage, taler-
act ddp In this step, we set tolcmnc-that is, a range of admissible value*
around the target v ~ l u e aof the control parameten identified in the parameter design
phase. Wc do this only if tht performance variatio? achiwed by the settings identified
in tht paramatcr design stag8 is not acceptrbk. For instance, in the example concern-
Ing the -diameter of coIlaplble rubes, tolcr&n#s for the mold pressure could be
xi f Axl, or thoat br the pellet size could be so* AQ, where Ax1 and Aso represent the
permissible variabilities br mold pressure and pcllct size, respectively.
Toleraneer that are too rigid or tight.wil1increlua manufacturing costs, and tolcr-
once8 that art too wide will incrtaw parformanw varirtbn, which in turn will increase
the &tomtrb loas. Our objcctiva is to find the bptimal trode-ulf between.these two
fortr. Usually, after the parameter dosign stage is completed and the parameter con-
trol wtdnp,.are adjusted, temflmmtlon (or w M w ) e q d m t n t r are conduclcd.
Them ekpettmtnts reltfRrm the degrcc of improvement realized when the chosen pn-
, ramettr settin ate w d . It is possible thnt the observed degree of variability of the
, $
performarno aracterhtie is more than what we have in mihd. I i this is the cnse, toler-
a m d&@ may be nscdcd to reduce the variability of the input pnramcten. whieh
, . 411thereby reduce output vnriation.
cal quuatlonn'ntthis stngc concern which pnrnmotcn arc to Ix: tightly con.
%
ITBW. ir always a factor here. I n most cnsea, the costs or tightening the vnrinbility
of mtml -on differ, Here sgain, a t r a d e 4 will be made on the tlltcl of n pnramc-
, tor o n , boutput ehareeteriudc and the cmt of tightening it to desirable bounds. ' ,

Aa noted &eartier, h the 'h#uchI philosophy, quality is tho lobs imparted to society frdm
tbr time a product b shlppd. I n t h preceding section we extended this definition to
include the loas incurred whilo the product is being manufactured. The compancnts of
1- jnclude the eqxmu, waste, and lwt opportunity that result when a product rails to
meet the t-t valw exactly, During production,,ewts suCh an inspection,scrsp, and
rswotk cantribute to loss. Although thdw costs Ira sary to account for, them are oth-
era that are much more diPtlcult to meawe, such an the tm associated with customer
dlwathfadon t h ~ arises
t fmm varlation In product8 and services.
Proflt-Weaning mensum aueh as productivity improvprnent end waste reduc-
don MIm qtant, but they a= bounded by weir factors as labor und material costs
and techno& Real y a n h in market s h w , however, is very much influenced by so-
dctythnt 11, b;y+customcrsatirfacdon. Price and quality arr!critical bctors in this con-
text. An cost, mcurred by the cwtmer or any losg muItlng from poor quality wiH
&
haw I Mesnt aegallw impact. On the other hand, a satisfied customer who real-
tea savinpiFom uring the produet or stmce,will turn f over many times to grcntty
improve rnarkst l a r e , Thew I h r lend us to conclude that varinbility is the kcy con-
#pt that &tea product quality to dollan. Wc can express this variability lhrough a
low iclnctton.
Th& qullity lma function ir stated in financial terms, whieh p ~ ~ a common
~ i d ~
langmp lor vrriour entitior within an organiz~lion,such as management, engineering,
and pt4duction. 11 can also be related to performmoe mcasurcs auch as the signal-t*
now ntlo, which i~ wed In the parameter dcaign phase; The loss function is assumed
,to be proportionnt to the squam of t k deviation of the qunlity chnractcristic f m the
tnqct wlue.The traditional notton or the Ims hnction is that as long as the pmduct's
qualtty characteristic is within artain spaifimtlon limits or tdemnces. rio loss is in-
euhad.Outside thesespeeilkatima, tha loss ttlkee the form of n step function with a
anatant valuc A. Figure 12-14 shows thin tradttimaI loes function.
Thore are Jrawbeks inherent to thin notion of 1- function, ObviaJly, thc tune-

or j ~ nbWet
r
tton makes no distinction b e t w n a product whas~characteristic is exacti an target
(st m) versus me whose ehrractwistic Is just below the upper specillcation I mit (USL)
the lower apeeidcati~nHhft (ISL).Thtre will be parformnee differ-
trim in thew product#,but the tradilonal loss lunction does not reflect these differ-
earn. Furthemore, it is diflklt to jjwtfy the ludden step inaease In lost ,as the
ualily chnrnctedstic just axcucda the spcciflcrrtlvn limits. Is them a p t functional
&fenme bonnen a product with a quality chamcter(~tisv a l w of U9L d and one-
with a value of USL + 6, where S is very small and approaches zero? Most likely,the
answer ia no. Additionally, It is unreasonable to say that the law remains constant, at 8
vJua A, far all valucs of thc cl~aracteristicbybnd ths specifidan limits. There will
obvioudy be tr 8ignilbnt functional difcrunca between productl that Arc barely above
the USL and t h e e that are far above the USL. 7%- differences would cause a per-
forman* vnrhtion, w the laas to r d e t y would be different.
The *hguehi loa function mercornm thesc dehtcncies, because the low in-
c r o w quadratically with inmasing deviation from thc target value. In the following
a discbm expressions for the lafunctions based on three situations: tar-
s u ~ t i o n we
get is &st, s d l a r L tKner, and larger is better.

Lets consider charactcrfstia fot which a target (or nominal) value is appropriate. Aa
tho quality eharncteriaic deviatu from the target value, t& loss increases in a qua-
dntlc manner, as ahown in Rgure 12-11. Examples of quality Eharacteristim in this EPP
c p y Include product dlmcdw such as length, thifknass, and diameter; a product
Aaracterirtle such as the W t y of an oil; md a serviw chamcteriadc such aa the da-
grea of mm%g~rnent fnvolwmsat. Tbguehik lora function Is given by
L(y)=k(y-m)' Ilm
whws k Is a proportionhllly matant, m is the target valuo, and y is the vaiuc of the
q d t y chuactarlstic Note that when y m-that is, when the value of the qudity
chfitactsdtfc is on urpt-+he 1- Is wm.
n r eonatant k can ba evaluated if tho 1- LCy) Is known for any particular valw
of the q d t y cbaractetistiq It is inflwnad by the financial importance of the qualify
chlvactd8tk For itrrtanw, if n Critical dimension of the braklng mechanism in an au-
tomobib &vimiK#n a target value, the 1- m a y incream dramatically. Figurc 12-15
Iharm three bhidm8 with d i f f c ~ dogre&t of financial importance. Thc
8t- the w,tba mors Lamant the 109s function. In Figure 12-15. loss function I
A is mom m t than loas futictions B and C. Ti
'Ib detm&m the d u o bf the constant k, suppose the functional tolerance range of
the @ty -ie -
is (m 4rn +A), as shown.inF p 12-11. Thisrapresentsthe
bh variation, beyond which the average product d a s not function
rage customer vvlewpoint is wpmhnted. Sup- thu cunsumcr$
a-h 1SA when tho quality CAaracterhtic is at tha IhIt of the functianal toleratfec.
Thb b represtntseasb to the comwner for repair or wplmmnt of the praduct, with
the Prwciatedd h t l l r d u n , Using oq, (1269),we find rf# p-lity m t a n t k as

given by q,(12.69)
The lw'funet~an can b e rcwdrtin as

The e r p c h d bw II the mean la= over many instances of the product. The
eqxctatiop i s taken with r c s p t to ChC distribution of tho quality characteriatie. )!
we have ,

Hem, M!SD mpmwntr the meah quwe &rlahn of y a d is e8timabd as

owt n.mmpIe d n Item, ln this oxprcssion; ~ # r C v represent


) the mean rmd vnri-
ance of y, ~ p d w l yNote -
, that MSDCV) = Vat01 + @ mI2: that is, the mean square
dovledon of a quality chatactsrlutb is the sum of the variance of the characteristicand
the squared Maa,
WIth the objqive of hdnIraIalng the expected losr, note that ifthe menn of the
charactodsttc Is at the target vdqs m, the mean square deviation is just the variance of
The qumtity E{lly)can bo admated from a sample of H itmu as
2 lly?
-
L
1
n
-me 12-9: b t a m m t m i e r s m for the bight ot n staarlng h n l m ere 15 f: 0.m
m, For a prduct fiat jwt errcscda thew IImlM, the Eaa to the c~6tontcrfdr getrfnf i t fixed b SSQ.
Ten m u ch am randmly mlmtd and ald the following heights (in meters): 1,93,1.49,1,#,
1.49,2~1S;1,1~,153,t51,nndIJ2. 8
ndIhervcnpbporpmdwtltam.
Soldon. Note that W h a ahuation In which ~ g skt best. 'ha mt wluem is 15 m.
me lm function b @wn

1 where y rcpmcna the height of the #teering mochrnlm and k k a ptoponlwlnlity mutant.
OknthaInIarmafian

Tho Im function is
-?.s)*
r.IY) 1 x 4 , ~
The expected Iws per item is given by
Qtw] I25,mEO- tgr
-
w h m Eb 1 ~h artlmrtcd
) ~ nf o l m

1I Hcncc, tho expected Iwr per ltom h

b m p h 12-70: Refor to Example 12-9 mneerning thc hi$ht of Inberlng &nf rn. 7'bu
m~ufaohnwfm d d e t i n g changing tbc ptodudon to reduce tbc varir8mty in tho wt-
plt. The addjtlmal ~ o r ftor he new p r o w is enrirnated to h S.Mlltom. The annual
,
tion ia 20.000 items Eight Items are randomly selected frob the nuw procum. yicl ing the
following bdght.: 151,l,M,1.49,1.52,1.52,1.50, I,&. 1.51, Is the new p
?
m car-ePRdant7 If
so,what is the annual savings?
S d d h The las lunctinn fmm Exrmplc 12-9 ia
to)= 1I t 5 , r n - 1.5y
1 For the new p-
d
we s ~ t i a t sthe mean q u a , d c v ( d a m v n d h e mr8a value of?.%

Tba axpcatd Id pr item for tho former proccsn 'b%12S, which W t a a Wap of
56I.25 - =A4 n $37,81litam, md mk ddad cm b only JS.3Wltcm. Thw, tb nst navln@ per

i I
, itan by udnp ths new -
b $3761 $5.50 n $3231, makihg It eoFt-dMmt.
The net mnrulrrvhga compaiod to the famrar pmem ia 20,000(3231) = WW.

You might also like