隨機最佳化之反應曲面法架構

Metamodel-based Frameworks for Stochastic Optimization

謝岳峰(Hsieh, Liam Yuehfeng)

指導教授 ： 張國浩

國立清華大學/工學院/工業工程與工程管理學系/博士(2013年)

摘要

Metamodel作為物理系統或模擬模型的替代模型，可用來協助描述與了解複雜的隨機系統之輸出入關係。在眾多建構超模型的方法當中，反應曲面法(Response Surface Methodology, RSM)為最廣為人知的優秀技術。此方法基於統計實驗設計的基礎，使其能夠有效率地建構出可靠的Metamodel。當Metamodel具有良好的代表性時，能夠提供有意義的資訊來幫助了解所欲研究的系統，並且進一步地協助系統最佳化的進行；此類以基於反應曲面之隨機最佳化演算法架構被稱為Metamodel-based Optimization (MBO)。 不論是隨機環境下的物理實驗或是模擬實驗皆能夠運用MBO來協助系統最佳化，能夠省去大量的運算時間是此類方法的主要優點。本研究基於一個MBO的概念模型，發展三種以RSM為基礎之隨機最佳化方法，分別用來協助處理不同類型的隨機最佳化問題。最後藉由在實證問題上的應用，藉此檢視所提出方法之效果與實用性。

關鍵字

反應曲面法 ； 隨機最佳化 ； 超模型 ； 模內裝飾技術 ； 半導體製造

並列摘要

A metamodel is a surrogate model of physical processes or simulation models, and is used to represent the input-output relationships of complicated systems. Metamodeling represents the process of constructing a metamodel, and Response Surface Methodology (RSM) is one of the most well-known techniques used to produce a metamodel. RSM has some inherent advantages over other metamodeling techniques due to statistical experimental design fundamentals, that made this technique more effective and reliable. The fidelity of a metamodel provides useful insights to understand parameters of interest in a system and assists in system optimization, and this technique is known as metamodel-based optimization (MBO). MBO is applicable to both physical experiments and simulation experiments. Based on a conceptual MBO model, we introduce three RSM-based frameworks that allow for efficient development of a metamodel based on variables of interest that supports MBO in certain cases. In our research, three empirical studies have been used to validate the viability of proposed frameworks among physical experiments and simulation experiments, respectively. Finally, we provide the conclusion and describe future research topics.

並列關鍵字

Response surface methodology ； Metamodel ； Stochastic optimization ； In-Mold Decoration ； Semiconductor manufacturing

參考文獻

Chang, K.H., Y.H. Huang, S.P. Yang. 2014b. Vehicle

Altan, M. 2010. Reducing shrinkage in injection moldings via the taguchi anova and neural network method. Materials and Design 31 599-604.

Arlot, S., A. Celisse. 2010. A survey of cross-validation procedures for model selection.

Backus, P., M. Janakiram, S. Mowzoon, G.C. Runger, A. Bhargava. 2006. Factory cycle-time prediction with a data-mining approach. IEEE Transactions on Semiconductor Manufac-turing 19(2) 252-258.

Ball, P.D., S. Roberts, A. Natalicchio, C. Scorzafave. 2011. Modelling production ramp-up of engineering products. Proceedings of the Institution of Mechanical Engineers, Part B: Journal of Engineering Manufacture 225(6) 959-971.