Plackett-Burman design of experiments, in a nutshell, is a technique used to identify the important factors in the early stages of experimentation. This technique works best when two factor interactions are assumed to be trivial.
Full Factorial designs and Plackett-Burman designs provide crucial information about the first-order models and, when center points are included, can also show the existence of curvatures.
An experiment was conducted to compare the two designs and prove their individual effectiveness. The design of full factorial experiments took 5 factors (A, B, C, D, and E), each having 2 levels (1 and -1). When plotted on to a table, they revealed a characteristic of response (Y). A normal probability plot then showed the significant factors and the interactions between them.
Next, the main effects plot showed factors which did not show a horizontal line indicate a significant factor. This in addition to the ANOVA, confirmed that factors B, C and D are the crucial factors.
Therefore to change the nature of response and make Y greater, the levels of the insignificant factors were changed to +1.
When the Plackett-Burman techniques were chosen, the same 5 factors were considered, and it was implied that the greater the value of Y, the better. Although, the design of this experiment did not show the interaction between the factors, but, it did confirm that factors B, C and D were significant.
Once both the experiments were completed, a comparative analysis revealed that the Plackett-Burman design required 12 experiments as compared to 32 in the full factorial design. Also, when full factorial was able to show the interactions between the factors, implementing the Plackett-Burman design increased the Y factor by a relatively small number (from 114.07 to 116.46).
To conclude, with their own little differences, using these screening techniques will depend on what the project demands and also on how much the characteristic of response needs to go up the analysis scale.
See full story on isixsigma.com