DOE planning

Guidelines

• The first question that must be answered is which type of DOE to use:
  • When setting the levels of the factors, you should limit the number of levels for each factor to two. DOEs with two levels for each factor are far more efficient (fewer experimental runs) than those in which some factors have more than two levels.
  • 2K factorial DOEs provide a very easy-to-use prediction equation, while General Full Factorial (GFF) DOEs do not. If obtaining a prediction equation is one of the important goals of the experiment, the 2K Factorial DOEs are highly recommended.
  • If any factor must be evaluated at more than two levels, use the GFF.
  • If you are using a 2K factorial DOE, consider using high-resolution (resolution V or higher) fractional DOEs instead of full-factorial designs. If you use a fractional DOE you can either reduce the size of the experiment or replicate the design with the same number of runs you would have had in the full-factorial design. If 3-way interactions are deemed unlikely and unimportant, the 2K fractional factorial design with a minimum resolution V is the preferred design because it will reduce the sample size while still providing the needed level of analysis.
  • For three factors, all at two levels, the 2K full factorial in eight runs is generally recommended.
  • For four factors, all at two levels, the 2K full factorial in 16 runs is generally recommended, unless the runs are extremely expensive and it is not important to estimate all 2-factor interactions.
  • For five or more factors, all at two levels, 2K fractional factorial designs with resolution V or higher are generally recommended.
  • Use the design-selection tables in Minitab to help you select an appropriate DOE (for example, the number of runs, given the number of factors you want to include).
• Whenever possible, the runs in the experiment should be done in random order to prevent confusing a factor effect with the effect of an untested factor (sometimes called a lurking variable).
• Replicating the DOE increases your ability to detect smaller factor effects. If you are not sure how many replicates you need, use the power and sample size calculations in Minitab to determine the number. To use this command, you will need to know (at least approximately) the standard deviation of the output, the size of the effect you want to detect, how much risk you are willing to assume at both missing an effect of interest and wrongly determining an effect is statistically significant.
• The 2K factorial DOE (both full-factorial and fractional-factorial) relies on the assumption that the effects of the factors on the response are reasonably linear (can be modeled adequately with a straight line) in the inference space. You should include center points in your 2K factorial DOE whenever you doubt the linearity of the effects. The center points produce a test for curvature, that is, they test the assumption of linearity. If the curvature is statistically significant, you must still decide whether the amount of curvature present is of concern from a practical standpoint.
• When adding center points to the DOE, the following procedures are often recommended:
  • Use the current process settings of the factors as the center point, which gives the operators running the experiment a comfort level with familiar settings of the factors.
  • Do not fully randomize the center points in the DOE. Instead, put one or two center points at the start of the experiment, one or two in the middle, and one or two at the end, which provides a check for trends during the running of the experiment.

How-to

For more information, go to Insert an analysis capture tool.