This thesis compares between three alternative experimental designs as to their applicability to correctly identify the component variations that have a statistically significant effect on product performance. The comparison is based on the number of correctly identified independent variables, number of component parts required, and number of experiments required; fewer parts and experiments is better. The three experimental designs are standard fractional factorial, double fractional factorial (consisting of two standard fractional factorials called the positive and the negative fraction), and Taguchi's inner and outer arrays.
To determine which design can correctly identify the significant component variations requires a case study where the correct answer can be determined. This is possible if the case study can be simulated. The case study selected was an idler wheel. Thus, a simulation was used to evaluate 3 experimental designs, which if successful, could be applied to other case studies where simulation cannot be used. The simulation program was 3Dcs (Dimensional Control Systems Inc., 1996). It was used not only to determine the correct sources of variation, but also to conduct each experiment instead of constructing physical prototypes.
The results from comparing the three experimental designs clearly show the standard fractional factorial to be superior to the other designs both in terms of determining the significant variables as well as doing so with fewer parts and runs. The double fractional factorial did not prove to be particularly useful, and Taguchi's required too many parts and runs and did not provide superior results. Thus, it is not recommended for tolerance design.