Candidate Models for Modeling Quality Characteristics in Process Capability Analysis

Capability analysis of manufacturing processes needs to appropriately model quality characteristics (QCs). The sample data for modeling can be divided into three categories: (a) original measurement data of QCs, (b) transformed or normalized measurement data, and (c) data of a composite QC obtained through aggregating several QCs. Generally, different categories of data require different types of distribution models. The purpose of this paper is to identify potential candidate distributions for the first and second categories of data. This is done through fitting seven realworld datasets to eight candidate distributions, which can be divided into two types: (a) common distributions like the Weibull distribution, and (b) the folded-normal distribution (FND) and its variants such as generalized and power halfnormal distributions. Main conclusions are that the normal assumption does not hold for the data considered in this paper, the measurement data can be appropriately modeled by some of the common distributions, and the transformed data can be appropriately modeled by the FND or its variants. These are useful for QC modeling in process capability analysis.