Roslund, Jerry L. (1989, ASQC) Saturn Corp., Troy, MI
Factorial experiments are powerful statistical methods to find the optimal combinations of several factors, especially when interactions are present. Fractional factorials are economical because three-factor and higher order interactions are traded for knowledge of more factors. However, this can lead to confounding difficulties that distort our view of the main effects. This paper describes a novel approach for designing factorial experiments using preferred columns in Taguchi-ordered orthogonal arrays.
Preferred Columns give the highest resolution (least confounding) for a given number of factors in an array. They are identified by geometric symbols that correspond to the number of factors being studied. This method greatly simplifies the setup of experiments because you only need to count the number of factors and then assign them to the Preferred Columns. The preferred columns cover designs from a full factorial to a saturated design depending on how many factors are loaded into the array.