實驗計劃中受限資料的非參數分析模式

Abstract

在工業界中常須進行實驗改善產品，而在實驗進行中，有很多狀況下無法蒐集完整的實驗數據，例如：實驗進行中突然斷電、儀器的限制或為了節省實驗成本等，這些狀況下都只能蒐集一定範圍內的數據，這種資料稱之為『受限資料』(CENSORED DATA)。受限資料所包含的因子資訊較少，作資料分析時特別困難，也因此須特別小心。 目前所發展出來的方法大多未考慮因子的變異性，且大多數的方法又太複雜，為了改善這些方法的缺點，本研究乃利用迴歸分析及非參數等方法，構建一演算法來分析多因子及多水準的受限資料，以找出最佳的因子水準組合，最後並以實例來作分析，以驗證本研究所構建的演算法之有效性及可行性，此法經本研究驗證證實具有相當的有效性，能使受限資料分析過程序簡化不少。

Experimental design is a critically important tool in the engineering world for improving the performance of a manufacturing process. It also has extensive applications in the development of new processes. Therefore, engineers usually improve the quality of products through the experiments. However, sometimes due to some controllable or uncontrollable causes (such as the damage of the instrument, out of electricity during the experiment, limitation of time and cost, etc.), only part of the experiment can be completed. In this case, the result of the experiment consists of the "complete" data and the "incomplete" data. Such incomplete data is called "censored data". For example, in the reliability testing for the I.C. products, we often obtain censored data from the experiment. Since the censored data contain less information than the complete data, it is more difficult to do the analysis. The objective of this research is to develop a cost-effective quality improvement technique for analyzing the censored data. Hopefully, through the correct analysis of the censored data, rather than redo the experiment to obtain the complete data, we may be able to obtain some important information about the optimization of the production and process. Since the procedure of nonparametric method is much easier to be understanded by the engineer, so the technique we plan to develop in this research is to analyze the mud-factor and multi-level censored data using the regression analysis and nonparametric method.