以動態規劃法進行TFT-LCD排序機派工之研究

Abstract

薄膜液晶顯示器(TFT-LCD)的生產主要有四個製程：陣列製程、彩色濾光片(CF)製程、組立製程及模組製程。其中組立製程生產TFT基板，而彩色濾光片製程生產的CF基板，將一片TFT基板和CF基板壓合後便形成TFT-LCD基板；為了方便生產現場的運送，TFT基板和CF基板是由卡匣承載，一個卡匣約可承載10至20片基板。在重視良率的TFT-LCD產業，TFT-LCD廠在生產時會面臨兩個決策問題，分別是基板配對決策及排序機派工決策，基板配對決策是將TFT基板和CF基板進行適當的配對來最大化產出良率，此決策問題在過去的研究中已有最佳演算法；在進行基板配對決策後，一組配對的TFT/CF基板可能分散在不同的卡匣中，因此需利用排序機對基板進行抽取重置(pick-and-replace)作業，以形成一個新卡匣。在生產環境中，有效率的排序機派工法則應能最大化下游組立製程的產能利用率，因此本研究將排序機派工決策模組化成動態規劃問題進行求解，在實例驗証後得到的結果指出，本研究所發展的動態規劃演算法顯著優於過去研究的啟發式解法。

The manufacturing of TFT-LCD involves four main processes: array, color filter (CF), cell, and assembly processes. The array process produces TFT plates while the CF process produces CF plates, where the two processes transport plates in a container is called a cassette that contains about 10-20 plates. A TFT-LCD plate is the combination of a TFT plate and a CF plate. In a low yield environment, a TFT-LCD factory faces two decisions: plate matching and sorter dispatching. The plate mapping decision, which has been well solved in literature, examines how to map TFT plates to CF plates to maximize the resulting yield. Given an optimal plate mapping, a pair of TFT/CF plates may be located in different cassettes and requires a sorter to pick-and-replace the plates to form a new cassette. Practitioners report that an effective sorter-dispatching algorithm is needed to enhance the utilization of the downstream cell process. Other than using heuristic rules, this research uniquely models and solves the sorter-dispatching problem as a dynamic program. Numerical experiments indicate that the DP model significantly outperform the prior heuristic method.