Page 509 - Design for Six Sigma a Roadmap for Product Development

468 Chapter Twelve

From Fig. 12.23 we can clearly see that the elementary effect method
(method 2) and response surface method (method 3) perform consis-
tently with low relative error measured by NED. The stepwise regres-
sion method (method 4) and Draper and Stoneman method (method 1)
displayed larger relative errors.

12.7 Summary

1. There are two main bodies of knowledge in DOE, experimental
design and experimental data analysis.
2. Two types of experimental design strategy are discussed in this
chapter, full factorial and fractional factorial. A full factorial design
can obtain more information from the experiment, but the size of
experiment will grow exponentially with the number of experiment
factors and levels. A fractional factorial design obtains less infor-
mation from the experiment, but its experiment size will grow
much slower than that of full factorial. In addition, we can adjust
the resolution of fractional factorial design so it can obtain needed
information while keeping experimental to a manageable size.
Therefore, fractional factorial design becomes the “workhorse” of
DOE in the industrial application.
3. The main DOE data analysis tools include analysis of variance
(ANOVA), empirical model building, and main effects and interac-
tion chart. ANOVA is able to identify the set of significant factors
and interactions, and to rank the relative importance of each effect
and interaction in terms of their effect on process output. The
empirical model, main effect chart, and interaction chart show the
empirical relationship between process output and process factors,
and they can also be used to identify optimal factor level settings
and corresponding optimal process performance level.
4. In industrial application of DOE, it is quite possible that we cannot
complete the full DOE; it is either full factorial or fractional factor-
ial experiments, due to various reasons. In this case, there will be
missing data in the factorial experiment, and we call them incom-
plete DOE. There are several effective incomplete DOE data analy-
sis methods discussed in this chapter, and they can deliver quite
good results in data analysis.