利用自組性演算法與基因演算法於混合實驗最佳化之研究

Abstract

在許多產業(如：食品、化工等)，應使用混合實驗(Mixture Experiments)最佳化設計問題，不適用傳統的實驗設計(Design of Experiments, DOE)來規劃實驗以求最佳配方。此乃因混合實驗之反應變數是受因子(或成份)間之相對比例所影響，而一般實驗之反應變數是受因子之水準所影響，混合實驗之目的在於找出最佳成份比例配方，故傳統之實驗設計手法不適用於混合實驗。此外，隨著科技日益精進以及消費者對產品品質之要求越來越嚴苛，使得產品設計日趨複雜，產品之品質已非單一品質特性所能決定。因此本論文之主要目的是利用類神經網路中的自組性演算法 (Group Method of Data Handling) 及基因演算法 (Genetic Algorithm)，發展出一套同時最佳化單品質及多品質特性混合實驗之演算法，以求出一組可使產品品質特性最大化的成份比例。最後，本論文以一個文獻中之案例及台灣某公司汽車煞車用橡膠皮碗之實例，來驗證本論文所提出之多品質特性混合實驗問題之最佳化演算法確實有效可行。

In some specific areas, such as chemical or material experiments, engineers often misuse factorial design on mixture experiments. Because the responses of mixture experiments are affected by the proportional relationship among the factors (or components) rather than the quantities of the factors, the conventional designed of experiments techniques are not appropriate for the mixture experiments. Moreover, with the improvement of technology and the increasing demands from the consumers, product design is becoming more and more complicated. Optimization of a single response can no longer satisfy the needs of customers. Therefore, this study utilizes Group Method of Data Handling (GMDH) and Genetic Algorithm (GA) to develop a procedure for optimizing single response and multi-response mixture experiments. Two cases from previous studies and a real case of rubber bowl production from a Taiwanese automobile company are utilized to demonstrate the effectiveness of the proposed procedure