- 學位論文
運用模擬與類神經網路預估系統於彩色濾光片廠之投料研究
Order Releasing Using Simulation and Neural-Networks Estimation System for TFT-LCD Color Filter Plant
摘要
TFT-LCD(Thin Film Transistor Liquid Crystal Display)產業分成薄膜電晶體陣列製作(TFT Array)、液晶面板組裝(LC Cell Assembly)與液晶面板模組組裝(Module Assembly)三大製程。目前台灣TFT-LCD產業蓬勃發展，在如此激烈的產業競爭環境下，已有許多研究進行改善生產之週期時間(Cycle Time)、提升產能問題以及成本與利潤的權衡問題，除了業者在新技術上佔有先機外，如何在既有生產環境下有效運用有限資源及準確預估產量，即為產業所面臨的議題。 本研究提出一適用於TFT-LCD的產量預估系統。透過AutoMod模擬軟體，建構符合實際生產環境中的情況，根據生產環境中的可變動因素，可進行What-If觀察與分析，但因模擬過程需花費不少時間，因此結合類神經網路之方法，除具有準確度情況下，亦可大幅降低模擬過程所花費大量的時間。由於本研究欲探討之產量預估問題，一般傳統常利用試誤法(Try and Error)或經驗判斷來規劃實驗，但其方法效果有限，故藉由混合實驗(Mixture Experiments，ME)與多投料法則進行產量預估之實驗。 因此，本研究使用產量預估系統模擬預估實際工廠運作情形，並導入各種可多投料數量之實驗，比較各種情況下之總產量，發現在各種情境下，本研究方法皆能有效預估出實際總產量。然而，提高更多的投入量時，卻不一定能夠有更佳的產量，因此可知，本研究方法除了於各種調整參數水準情況下預估產量，且可提供改變原生產排程之投料比例，改由有效控制投料模式過與閒置時間，發現其可減少生產總投入量，仍可達到相似產量的情況。最後將預估結果與模擬結果做比較，驗證本研究所提出之產量預估系統確實有效可行。
並列摘要
TFT-LCD industry in Taiwan grows vigorously at present. Environment of intense competition in the industry, there are many studies to improve the production cycle time (Cycle Time), to enhance the capacity and to balance cost and profit trade-offs. Under such intense competitive situation, besides take the advantages on the latest technology, it also needed to know how to effectively use limited resources and to accurately estimate production in the inherent environment that will be the anxious issue. This study proposed an optimized production estimated system for TFT-LCD industry. Through an AutoMod simulation software, a simulated production system was constructed to match the real one. The What-If analysis was implemented to vary the factors in the production environment. Since simulation process takes time, this study integrated artificial neural network and simulation to develop an estimated system for throughput that can significantly reduce the time of simulation process. This study is to investigate the issue of production estimation. Traditionally, a trial and error method or experience judgment was applied to plan the experiment to get the best combination of factors. But the method limited its speed of calculation and affected the accuracy of experiment. Therefore, a mixture experiment (ME) and release methods was conducted to estimate the production in this study. This study estimated the throughput and simulated the actual operation of the plant situation. The throughput estimated system (TES) was effectively developed to estimate the total throughput. This study found the TES can reduce the production of the total initial scheduling, and still achieve the same total throughput between different conditions of factor. Finally, the results of TES compared with the results of simulation, and the throughput estimation system were verified that it is more feasible and effective.