基因演算法用於增加產能的研究-以TFT-LCD廠Array製程RGV搬送為例

Abstract

TFT-LCD廠主要分為三大製程：陣列(Array)、組立(Cell)、模組(Module)，其中陣列製程由於機台昂貴，最重視的是機台使用率及產能。本研究是針對陣列製程的工廠，此工廠會因為不同機台製程特性的不同，將工廠劃分成為數個區域，再針對區域配置適合的機台設備，並且每個區域內是由RGV(Rail Guided Vehicle)負責搬送。此工廠的特性是每個區域同一時間會有許多的搬送命令，而RGV一次執行一個搬送命令來完成區域內的搬送工作。因此，藉由調整區域中RGV搬送的順序，就可以影響區域整體的生產時間，進而提昇區域的產能。 而目前相關研究，一般而言皆只是提出一、兩個影響搬送效能的因素，找出改善小部分狀況的方法，如此未能考慮其他大部分的狀況，也沒有探討其他因素的影響。另外，這些研究皆是針對在AGV(Automated Guided Vehicle)搬送下，因此結果是針對AGV的特性，無法應用到其他特性的工廠中。 因此，本研究希望針對陣列製程工廠的單一特定區域，導入一適用於RGV特性的搬送策略來選擇搬送命令，進而提昇區域的產能。首先建立一個盡量符合現實工廠狀況的模擬器，並提出相關的假設，如每次離開此特定區域後再回到區域的時間。製作此模擬器主要是希望利用模擬器計算此特定區域的總生產時間，利用生產時間來評估區域產能的大小，時間愈短表區域產能越高。接下來提出搬送策略為命令屬性的線性組合。由於屬性之間的組合複雜且互相牽制，故利用基因演算法找出較適合的權重，得到一個針對此特定區域較適合的搬送策略，如此一來就可以藉由此搬送策略提昇區域產能。經由實驗的結果，不同的搬送策略對於同一個區域生產時間有顯著的影響，針對特定的區域可找到較適合的搬送策略，可以有效減少區域生產時間提昇區域產能。

As we know, TFT-LCD factory contains three major process：Array, Cell, and Module. Because the manufacturing equipments in Array are more expensive than others, we are very concerned with the throughput and the utilization rate of the equipments. TFT-LCD Array factory can be divided into several areas according to different process features of equipments, and then we will configure equipments and RGV (Rail Guided Vehicle) for transporting Lot between equipments for each area. The issue of factory is how can we choose one command from several transport commands in the area to improve transport efficiency since RGV only handles one command at a time. Therefore, the ordering of the commands can affect total manufacturing time and increase throughput in an area. In related researches, which are based on AGV transport and can only be applied for AGV behavior, only one or two factors influencing transport efficiency have been proposed. Hence, we hope to find an appropriate transport strategy to reorder transport commands for RGV in an Area of TFT-LCD Array Fab to increase throughput. First, we set up a simulator that is similar to real factory with some assumptions such as the time between Lot leaving the area and coming back to the area should be constant. Since simulator can calculate the manufacturing time of all Lots in the area, we thus evaluate throughput using the total manufacturing time where the less total manufacturing time indicates the throughput is high. Then we propose a transport strategy which is the linear combination of commands’ attributes, and then use GA to find more suitable weight and get one transport strategy in a specific area. Applying the obtained transport strategy to the specific area will increase throughput in this area. The experimental results show that difference transport strategy will affect manufacturing time obviously, and that specific transport strategy for specific area will reduce manufacturing time effectively and increase throughput in this area.