結合田口方法、類神經網路、期望函數與基因演算法於太陽能選擇性吸收膜連續濺鍍製程之參數設計最佳化

Combining Taguchi Methods, Neural Networks, Desirability Function and Genetic Algorithms for Optimizing the Parameter Design of Solar Energy Selective Absorption Film Continuous Sputtering Process

王繼青(Chi-Ching Wang)

指導教授 ： 蘇朝墩

國立清華大學/工學院/工業工程與工程管理學系/碩士(2011年)

摘要

在抗暖化聲勢如日中天之際，越來越多國家政府競相參與替代能源的研究發展。工業技術研究院團隊已對太陽能選擇性吸收膜濺鍍製程進行技術開發，在批次濺鍍技術的基礎上，建置國內首座捲對捲（Roll-to-Roll）連續式濺鍍製程原型機，並希望結合工業工程領域之相關知識，最佳化製程參數。
太陽能選擇性吸收膜連續濺鍍製程中，影響其成品品質的製程參數眾多，如果僅依據工程師經驗來判斷其設定值，可能導致製程不穩定而造成不良率提高。本研究針對太陽能選擇性吸收膜製程參數最佳化提出一系統性的求解程序，首先以歷史資料與工程知識篩選出重要參數，使用田口方法（Taguchi Methods）求取最佳化參數組合，另外也透過倒傳遞類神經網路（Back-Propagation Neural Network）建立回應值之SN比與控制因子間的關係模式，以期望函數（Desirability Function）合併輸出值作為適應函數，再利用基因演算法（Genetic Algorithms）求解最佳製程參數水準組合。
經由實驗驗證結果得知，本研究提出之方法能有效掌控問題核心，僅透過18筆資料與調整七個關鍵因子，就能使績效指標提升至國際級高品質標準內。能源節約效益相當於每年可抑制二氧化碳排放達27,553公噸，這是傳統烤漆技術的11倍之多，站在工業界的角度儼然是十分可觀的數字。

關鍵字

太陽能 ； 選擇性吸收膜 ； 連續濺鍍製程 ； 田口方法 ； 倒傳遞類神經網路 ； 期望函數 ； 基因演算法

並列摘要

“Anti-Global Warming” is a hot topic around the world. A growing number of countries have participated in the research of alternative energies. Industrial Technology Research Institute of Taiwan (ITRT) has already developed the technology in solar energy selective absorption film continuous sputtering process. The institute established the first Roll-to-Roll continuous sputtering process machine in Taiwan based on the batch sputtering process technology. ITRT attempts to combine the technology and the industrial engineering knowledge to optimize manufacturing process parameter settings.
For the extremely complicated solar energy selective absorption film continuous sputtering process, plenty of parameters would affect the output. If we only rely on engineer’s experience to determine the values, the defect rate may increase owing to the unstable manufacturing process. This study proposes a systematic procedure for parameter optimization of solar energy selective absorption film continuous sputtering process. First, historical data and engineering knowledge are used to determine the significant factors. Second, Taguchi methods are employed to find the optimal combination of parameters. Finally, back-propagation neural network (BPN), desirability function, and genetic algorithms (GAs) are utilized to obtain the optimal parameter level combination.
The experimental results present that the proposed method can control the core of the problem efficiently. By simply employing 18 data and adjusting 7 factors, this research can enhance performance metrics to an international high quality standard. Our proposed method can restrain 27,553 tons of carbon dioxide every year with respect to beneficial energy conservation, which is 11 times less emissions than what the traditional paint process produces. From the perspective of industry, this result is considerably impressive.

並列關鍵字

Solar Energy ； Selective Absorption Film ； Continuous Sputtering Process ； Taguchi Methods ； Back-Propagation Neural Network ； Desirability Function ； Genetic Algorithms

參考文獻

[46] 拓墣產業研究所，2007，全球替代能源的投資金額預測。

[1] Bowden, F. P. and Tabor, D., 1950, "The Friction and Lubrication of Solids" (London: Oxford University Press).

[2] Chou, C.-J., Yu, F.-J. and Su, C.-T., 2008, "Combining Neural Networks and Genetic Algorithms for Optimizing the Parameter Design of Inter-Metal Dielectric Layer," Management Track within WiCOM: Engineering, Services and Knowledge Management, Dalian, China, pp. 1-4.

[3] Cook, D. F., Ragsdale, C. T. and Major, R. L., 2000, "Combining a Neural Network with a Genetic Algorithm for Process Parameter Optimization," Engineering Application of Artificial Intelligence, Vol. 13, Issue. 4, pp. 391-396.

[4] Derringer, G. C. and Suich, R., 1980, "Simultaneous Optimization of Several Response Variables," Journal of Quality Technology, Vol. 12, pp. 214-219.