Interval Robust Design on Quality Improvement for Non-Normal and Contaminated Responses

Abstract

The basis of robust parameter design is the creation of a design that can resist the negative effects caused by uncontrollable or difficult-to-control external and environmental factors, which affect the product parameters in achieving product design during product realization activities. Robustness is the ability of a product or process to be least affected by variabilities caused by external factors. The success of the response surface methodology generally depends on a model chosen to fit the data distribution. Making incorrect assumptions regarding data distribution when creating response surface models can affect the effectiveness of the quality improvement strategy used. Non-normal or contaminated data is a common phenomenon in quality improvement applications. Although non-normal data is common in robust parameter applications, it is often the case that users ignore the underlying distribution shape of the data at the modeling stage and use normal theory techniques naively. This study proposes a dual response surface approach based on robust confidence intervals for cases where the experimental data do not meet normality assumptions or have contaminated data distribution. A new dual response surface methodology is proposed based on modeling the  confidence interval,  confidence interval, and  confidence interval formulations with the response surface methodology. All the proposed methods make the process median unbiased for the mean using the skewness of the experimental data. Two well-known experimental design studies are used to demonstrate the procedure and its advantages.