Page 457 - Design for Six Sigma a Roadmap for Product Development

416 Chapter Twelve

In this chapter, we are considering primarily full factorial designs
and fractional factorial designs. If the numbers of factors and levels
are given, then a full factorial experiment will need more experimen-
tal runs, thus becoming more costly, but it will also provide more infor-
mation about the process under study. The fractional factorial will
need a smaller number of runs, thus costing less, but it will also pro-
vide less information about the process.
We will discuss how to choose a good experimental design in subse-
quent sections.

Step 5: Perform the experiment
When running the experiment, we must pay attention to the following:
■ Check performance of gauges and/or measurement devices first.
■ Check that all planned runs are feasible.
■ Watch out for process drifts and shifts during the run.
■ Avoid unplanned changes (e.g., swap operators at halfway point).
■ Allow some time (and backup material) for unexpected events.
■ Obtain buy-in from all parties involved.
■ Preserve all the raw data.
■ Record everything that happens.
■ Reset equipment to its original state after the experiment.

Step 6: Analysis of DOE data
Statistical methods will be used in data analysis. A major portion of
this chapter discusses how to analyze the data from a statistically
designed experiment.
From the analysis of experimental data, we are able to obtain the
following results:
1. Identification of significant and insignificant effects and interac-
tions. Not all the factors are the same in terms of their effects on the out-
put. When you change the level of a factor, if its impact on the response
is relatively small in comparison with inherited experimental variation
due to uncontrollable factors and experimental error, then this factor
might be insignificant. Otherwise, if a factor has a major impact on the
response, then it might be a significant factor. Sometimes, two or more
factors may interact, in which case their effects on the output will be
complex. We will discuss interaction in subsequent sections. However, it
is also possible that none of the experimental factors are found to be sig-
nificant, in which case the experiment is inconclusive. This situation