A Taguchi experiment was designed to analyse the influence of seven factors (identified as Factor 1 to Factor 7) on porosity. A porosity score was defined for each casting. Porosity score of zero and one is desired, two and three is tolerable and above four is unacceptable. Two settings - setting 1 and 2 were defined for each factor.

We will focus on Factors 5, 4 and 2 and count number of castings produced in desired, tolerable and unacceptable category for each factor and setting.

Taguchi experiments are generally very simple and data can be visually inspected for at least for discovering main effects.The pattern that is likely to improve your process will emerge by representing data in the confusion matrix form.

This approach may not work for production data when interactions among factors and responses are important.

Patterns for Factor 5 are very clear. Setting 1 produces more number of castings with Desired Response and less with Unacceptable Response in contrast with Setting 2.

Patterns for Factors 4 and 2 are weak and similar.

Whether its Taguchi analysis or Factorial Analysis, both methods identified Factor 4 as most important factor and indicated significant difference between Factor 4 and 2.

The results are counter intuitive.

Reason: Statistics applies known distributions to unknown problems. Its not based on pattern recognition. Hence, it is not reliable in discovering interactions.

MetaCause software is based on pattern recognition and hence its results are reliable and accurate whether its production data or Taguchi/DoE data.