Page 279 - Six Sigma Advanced Tools for Black Belts and Master Black Belts

3:5
Char Count= 0
August 31, 2006
JWBK119-17
264 Strategies for Experimentation under Operational Constraints
Factor
Weld time 17.2 (+)
3.9 (−)
x 6
Air pressure 2.9 (+)

x 5 2.7 (−)
Hold time 1.5 (+)
Down speed 1.0 (+)
Trigger pressure 0.9 (+)

Effect on weld strength
Figure 17.1 Results of a 2 5−2 welding experiment.

x 7 assigned to air pressure, hold time, down speed, weld time and trigger pressure,
respectively. With the full data set, the effects of the five factors are summarized in a
Pareto plot in Figure 17.1. It is evident that even allowing for the existence of some
interaction effects, weld time stands out as the most critical factor in affecting weld
strength. The problem was subsequently solved via an upward adjustment of weld
time from the prevalent set point.
With these results, it could be reasoned that had the investigators been requested
to reduce the amount of destructive tests with the high-cost samples, one of the eight
experimental runs could perhaps have been omitted. For example, if the last run
were not performed, then either one of the effects associated with x 5 and x 6 could be
assumed zero, and, by virtue of (18.6), each of the estimated effects of the five factors
would be altered by 2.7 and 3.9, respectively. This would not affect the conclusions
of this screening experiment at all, since with the new numerical results, weld time
would still reveal itself as the predominant factor in a Pareto analysis. In such a
situation, the quality of decisions made in a screening experiment is seen to depend
much more on the relative strengths of the factor effects than on the completeness of
the experimental data set.

17.8 CONCLUDING REMARKS

Inasmuch as textbook conditions cannot be always satisfied in industrial situations,
planning of experiments should be done more in the spirit of adaptive design rather
than by unimaginative adoption of recipes. It should be pointed out that an ability to
deviate from the standard orthogonal arrays of Taguchi textbooks would considerably
broaden the application of Taguchi methods, a fact yet to be recognized by many of
the methods’ advocates. Non-Taguchi types who know enough to evolve their own
fractional factorial designs would appreciate the possibility of ‘customizing’ a design
matrix to cope with real-life constraints. Coupled with technical judgment, suitable
sequential and incremental experimental design and analysis schemes 9−11 can help