使用剖面資料資訊提昇製程監控能力之研究

Abstract

傳統的統計製程管制(SPC)方法是直接監控有興趣的產品品質特性。由於現在科技發達，我們可以很容易的隨著時間點的改變紀錄一些與產品品質特性有相關的量測值，而這些量測值可以提供產品品質特性額外的資訊。舉例來說，我們可以在產品完成前，在不同的時間點上記錄品質特性的反應變數值。若我們把這些記錄值視為產品的剖面(profile)資料，則最終產品之品質特性即為剖面資料的最後一點。我們可以預期這些輔助的資訊可以提供我們更有效的製程監控方法。 本篇論文只探討第二階段的製程監控。我們提出的方法是用有母數、無母數迴歸的方法來配適產品的剖面資料，且利用配適出來之剖面的最後一點來對製程進行監控。我們證明利用了輔助資訊來對製程進行監控會有較好的偵測力。

Traditional control charting methods in statistical process control monitor the quality characteristic of the product or process of interest directly. Now thanks to well-developed technologies, we can easily record many other related data that may provide additional information about the quality characteristic at different time points. For example, we can record the quality response variable along the time before the end product is finished. If we regard the record of these values as the profile of a product item, then the quality characteristic of the end product is the endpoint of the profile. It is natural to expect that this auxiliary information would be able to help enhancing the control process. In this study, we focus on Phase II monitoring. Our approach is to fit each profile by parametric or nonparametric regression methods, and then monitor the fitted endpoint value instead of the endpoint response. It can be shown that, with the additional profiles information, the fitted endpoint response has smaller variance than the endpoint response itself. Accordingly, when compared to the traditional approach, better detecting power can be achieved with the proposed approach