| Summary: | 碩士 === 輔仁大學 === 統計資訊學系應用統計碩士班 === 100 === Fraction nonconforming, which follows a binomial distribution, is one of most important quality characteristics for attribute processes. In addition, because of a great diversity of the quality characteristics of the products, the multivariate binomial process plays a major role in industries. The way of process improvement, when the disturbances have introduced into a multivariate binomial process (MBP), is an important research issue. This study considers two ways to improve the MBP, and they include the identifications of change point and sources of fraction nonconforming shifts. The maximum likelihood estimate (MLE) is one of the widely used methods to estimate the change point of a process. However, the MLE is criticized by its strict assumption of prior known process distribution and the derivation difficulties for a multivariate process. The MLE is not unique for a MBP since the mathematical form of a MBP is not unique. Additionally, a MBP is considered to be out of control when a signal is triggered by a multivariate statistical process control (MSPC) chart. However, it is very difficult to identify which quality characteristic (or characteristics) is responsible for the process fraction nonconforming shifts. This study is motivated to develop a mechanism to overcome the difficulties for identifying the change point and the sources of fraction nonconforming shifts in a MBP. The proposed mechanism includes the integration of two novel identification approaches with the artificial neural networks (ANN) and the support vector machine (SVM). The use and the superior performances of the proposed approaches will be demonstrated by conducting a series of experiments.
|
|---|
