A distribution-free method for full-factorial or fractionated reliability and yield trials

Product/process reliability and yield responses are often correlated variables. In pragmatic situations it is one of the two that is more conveniently measured. A distribution-free approach is proposed to treat the multifactorial screening/optimization problem of reliability or yield enhancement, regardless of encountering a full or fractional factorial experimental plan in either replicated or unreplicated form. Censored data are also handled. The nonparametric processor utilizes robust estimators and a reference law associated to Wilcoxon-Mann-Whitney order statistics. The structured model includes screening/optimization term arrangements to accommodate both main effects and their interactions. Four published case studies illustrate the method for real processes that extend to effect predictions up to three-factor interactions in reliability and yield datasets. The method appears to be competitively accurate to other methods, in spite of the fact that no other technique has been available to treat either reliability or yield data at the same time. The examined examples ensure the engagement of replicated and unreplicated multifactorial trials, permitting censoring, and stressing the presence of strong non-normal distributions in either the reliability or yield datasets. The versatility to switch from reliability to yield datasets is demonstrated without changing the theoretical framework of the data collection or analysis. The agility of the method is easily demonstratable, since it is iterative-free, simulation-free and resampling-free. The approach may be useful when rapid deployment and analysis of reliability research work is expected for product/process improvement due to economic and time limitations.