Page 293 - Six Sigma Advanced Tools for Black Belts and Master Black Belts

OTE/SPH
OTE/SPH
August 31, 2006
JWBK119-18
Taguchi Methods
278 3:6 Char Count= 0
37
out with Taguchi methods, see Tan Goh. ) In fact, for seeking and tracking optimal
parameter values during the very last stage of any optimization study, a capability
for specifying an experimental design matrix, mathematical modeling, optimization
and sensitivity analysis is absolutely essential, and cookbook procedures would not
suit the purpose at all. There have been a host of other objections and criticisms on
more theoretical grounds leveled by statisticians at Taguchi’s gamut of design and
analysis routines, 12,15,20,38−43 which will not be elaborated here. These, however, may
appear to be of somewhat remote concern from the point of view of certain quality
professionals who have come to appreciate a number of nontraditional applications
that Taguchi has developed and promoted for engineers; some such applications are
illustrated below.

18.5.4 Special applications
Taguchi methods stress a number of practical objectives beyond those associated with
the usual SQC or experimental design studies. These objectives are often related to
product or process development and design. Chief among them is the emphasis on
reducingvariabilityintheperformanceofaproductorprocess;this,asalreadypointed
out in the earlier discussion on problem formulation, is handled by using experimental
2
design to address a changing, rather than constant, value of σ of the response. When
2
input parameters are optimized for minimum σ , process capability (as expressed by
indices such as C p and C pk ) is maximized for a given operating technology, that is,
without additional capital investment.
Secondly, by introducing simulated noise parameters in experimental design, it
is possible to introduce adverse manufacturing conditions and environmental stress
into the study, thereby generating insight into product and process reliability in addi-
tion to quality which normally represents only stress-free and as-made characteristics.
A typical performance stabilization strategy is illustrated in Figure 18.7, where it is
shown that the performance index y is affected by the level of a design parameter
x d (e.g. material thickness) and that of a simulated environmental noise parameter x e
(e.g. temperature); x and x are low and high values, respectively, of x d used in a
+
−
d d
given experimental study, and x and x are simulated low and high values, respec-
−
+
e
e
tively, of x e . It is seen that if, as a result of the experimental design study, the design
y
+
x e
y *
−
x e

− + x
x d x
Figure 18.7 Parameter setting for robustness against noise.