Abstract

Successful designed experiments lead to increased process understanding. There are eight recommendations for success. First, have a well-defined purpose or objective for the study. Second, use a quantitative tool, such as a rating system, to measure responses. The measure should minimize reproducibility errors. Third, use knowledge about signal to noise ratios to determine how many runs are needed for a given statistical significance. Fourth, the run order of experiments should be randomized. Fifth, use blocking to weed out some expected variation. For example, day-to-day variation can be lessened by scheduling the runs into a different block each day. Sixth, when analyzing results, be aware of the effect of alias structure. Seventh, run the experiments as a series so that information obtained from one experiment is applicable to the next. For example, a screen design could precede a full factorial design. Eighth, before reporting the results, run a confirmation study to verify the findings. This paper uses a case study on variation of parts shrinkage to demonstrate each of the eight points.