Confidence Interval based Quality Improvement for Non-normal Responses

Abstract

Robust parameter design is an effective tool to determine the optimal operating conditions of a system.Because of its practicability and usefulness, the widespread applications of robust design techniquesprovide major quality improvements. The usual assumptions of robust parameter design are that normallydistributed experimental data and no contamination due to outliers. Optimizing an objective functionunder the normality assumption for a skewed data in dual-response modeling may result in misleading fitand operating conditions located far from the optimal values. This creates a chain of degradation in theproduction phase, e.g., poor quality products. This paper focuses on skewed experimental data. Theproposed approach is constructed on the confidence interval of the process mean which makes the systemmedian unbiased for the mean using the skewness information of the data. The response modeling of themidpoint of the interval is proposed as a location performance response. The main advantages of theproposed approach are that it gives a robust solution due to the skewed structure of the experimental datadistribution and does not need any transformation which causes any loss of information in estimation ofthe mean response. The procedure and the validity of the proposed approach are illustrated on a popularexample, the printing process study.