階層式多廠區生產鏈規劃之研究—以TFT-LCD產業為例

Hierarchical Multi-Site Production Chain Planning— A TFT-LCD Industry Case

陳盈彥(Yin-Yann Chen)

指導教授 ： 林則孟

國立清華大學/工學院/工業工程與工程管理學系/博士(2006年)

摘要

本研究是以TFT-LCD（thin film transistor-liquid crystal display）產業鏈面板製造商為例，探討階層式多廠區生產鏈規劃之問題。由於該產業的諸多特性或限制，例如各世代數工廠造成不同製程技術所衍生的產品不同加工路徑、薄膜電晶體基板與彩色濾光片基板的組立貼合形成供應網路流進流出量的不同、面板經濟切割數影響投入產出量的規劃、陣列製造廠瓶頸機台與副資源光罩之搭配限制等等特性，增加在進行階層式生產鏈規劃時的複雜度。
　　首先本論文回顧相關研究文獻，針對供應鏈多廠區的生產模式、各式生產鏈結構、相關規劃特性與限制等進行探討，提出多廠區規劃問題之分析架構，並以TFT-LCD面板製造產業為例說明先進規劃與排程模組系統。針對此產業階層式多廠區生產規劃問題本研究提出一兩階段規劃模式架構，包含TFT-LCD多廠區生產鏈之供應網路規劃模式，考量滿足最終產品需求數量與各產品族規劃的生產總量目標下，決定每個廠區未來每個月所需生產的產品數量及物料採購計畫；接著為陣列製造廠區日投入生產計畫之規劃並考量光罩和瓶頸機台配置問題。
　　本研究並以TFT-LCD產業為例，考量多限制式、產品族分解、規劃時間軸分解等特性，利用線性目標規劃方法求解供應網路規劃階段並進行分析探討，包含不同規劃決策模式之比較與相關成本之敏感度分析等。而陣列製造廠區日投入生產計畫問題，由於為複雜之NP-Hard問題，本研究發展具動態慣性權重與突變機制之二位元粒子群最佳化方法，有別於一般定值之慣性權重，本研究提出依演化代數之非線性慣性權重方法，與傳統標準的粒子群最佳化方法和遺傳演算法比較後，本方法搜尋所花費的時間相較其他方法來得迅速有效率，可快速收斂到近似最佳解處。

關鍵字

TFT-LCD ； 多廠區生產計畫 ； 粒子群最佳化

並列摘要

A hierarchical multi-site production chain planning framework, which consists of Supply Network Planning (SNP) and daily scheduling models, is presented in this study. Besides, a TFT-LCD manufacturer case in Taiwan is taken as an illustration to explain the planning issues.
Firstly, this article reviews the related abundant literature. An analytic framework for the multi-site planning problem based on the production modes, network structure, relevant planning characteristics and constraints is proposed. And then, we also developed the two-phase planning method for multi-site production planning problem. The first stage is Supply Network Planning, which decide the production amounts and material requirement planning among multiple plants in supply network. This problem is solved by preemptive linear goal programming considering the practical constraints, and discussed by the different planning modes and sensitivity analysis for the related cost factors. The second stage is daily scheduling in Array process, As a result of the complex problems, we develop the binary particle swarm optimization with dynamic nonlinear inertia weight and mutation mechanism. Compared with other methods, the proposed heuristic technique is verified for computation efficiency. It can converge to the best solution region very quickly.

並列關鍵字

TFT-LCD ； multi-site production planning ； particle swarm optimization

參考文獻

[1] Al-Fawzan, M.A., “An algorithm for production planning in a flexible production system”, Computers & Industrial Engineering 48, pp.681-691, 2005.

[2] American Production and Inventory Control Society, APICS Dictionary, 1998.

[3] Banerjee, A., “A joint economic lot size model for purchaser and vendor”, Decision Sciences 17, pp.292-311, 1986.

[5] Benjamin, J., “An analysis of mode choice for shippers in a constrained network with applications to just-in-time inventory”, Transportation Research B 24, pp.229-245, 1990.

[6] Berretta, R. and Rodrigues, L.F., “A memetic algorithm for a multistage capacitated lot-sizing problem”, International Journal of Production Economics 87, pp.67-81, 2004.