Multi-response quality improvement by non-parametric DOE methods

Quality management principles promote the systematic elimination of problematic areas of products before they reach the marketplace. It is a rare fact that a product is as good as optimising just a single quality trait of it. A practical and economical method is presented in this work for placing statistical significance on non-linear multiresponse-to-factor relationship during screening experimentation with unreplicated and saturated fractional factorial designs. A technique is developed on a simple additive rule for multiple ranked responses all of which are tested synchronously against a group of product or process parameters. The developed hypothesis is tested based on the multi-level nonparametric method proposed by Jonckheere-Terpstra. Advantages exhibited are the minimal computational effort and convenience in usage while relaxing strict normality assumptions. The theoretical developments are illuminated by a case study on a recently posed problem in the ever-popular area of small-component plastic injection moulding.