TFT-LCD陣列廠考量產品需求逐期變更環境下生產排程系統之構建

Abstract

生產系統在市場需求波動之環境下，各週期的產品組合發生改變時，容易發生排程穩定度及生產績效之變異，與傳統研究將產品組合假設為固定之情境不同，規劃上也較為複雜。故本文針對TFT-LCD陣列廠需求變動環境下建構一生產排程系統，包含主生產排程模組及生產績效估算模組，透過適當的投料規劃及負荷分配，以降低系統不穩定之負面狀況。 主生產排程模組之工作為決定每日投料種類、數量與順序，以及估算各產品層級流程時間。首先，將上一規劃週期之層級流程時間估算值設定為起始值，並依據期初在製品與投料工件所處層級，推算出預定於瓶頸工作站之負荷時點。接著，在產能限制下，以最小化每日負荷差異、投料量差異及未滿足需求為目標，求解出每日投料組合。將每日投料組合及期初在製品之數量作為輸入項，在考量製程規格能力、光罩數量限制及迴流次數下，分別以最小化機群及機台利用率差異為目標，進行各機群及機台之負荷分配，進而計算瓶頸機台之利用率。接著，透過等候理論中之Jackson network，求得非瓶頸工作站之利用率，以便套入Conway估算公式，計算工作站流程時間，並根據各產品流經各工作站之資訊加以估算得層級流程時間。當層級流程時間之估算值與設定起始值之誤差大於2%，則需將層級流程時間進行遞迴修正並重新規劃；反之，則執行生產績效估算模組。最後，於生產績效估算模組，估算需求逐期變動環境之各產品生產流程時間與最適在製品數量，以確保主生產排程得以落實。 根據本文案例之成效分析結果顯示，在投料法則為均勻負荷法下，生產流程時間之估算與模擬結果相比，平均誤差約為4.5%；而在固定在製品法下，平均誤差約為2%。此外，系統產出平均達成率皆約為100%。整體而言，本文所提出之生產排程系統，可應用於市場需求波動之環境下，並可將投料與生產績效規劃結果作為現場生產人員之參考依據。

When market demands changes periodically, the product mix in production system will be varied. This situation makes production planning more difficult since the stability of schedule is losing while the variance of performance is increasing. Therefore, this thesis constructs a production scheduling system for TFT-LCD array factory under demand change environment. To determine the daily release plan and to allocate the capacity of bottleneck workstation, the master production schedule module and production performance estimation module are included in the system. The duty of MPS module is to determine daily release type, quantity and sequence of products, and to estimate layer flow time of each product type. The initial value of layer flow time is set as that of last planning period. Based on this value, each lot either being work-in-process or planning to release will be projected the time arriving at bottleneck workstation according to its layer numbering at the beginning of a day. With such information, for minimizing the variation among daily release quantities, daily loading of the bottleneck workstation, as well as minimizing the unmet demands, an IP/LP model is built to solve the release plan under of capacity constraints. Then, base on the daily release amount and the quantity of initial WIP, the capacity load of bottleneck workstation is allocated to each machine group and then to each machine unit with consideration of the constraints related to process capability, quantity of available masks and the reentry times needed for each product type. IP/LP models are developed to minimize the variability among utilization rates of bottleneck machines. Given the utilization rate of bottleneck workstation, the utilization rate of each non-bottleneck workstation can be derives by solving Jackson network and the flow time of each workstation can be estimated by using Conway’s formula. Then, each layer flow time can be estimated so as to calculate the difference between it and its corresponding initial value. If the error is larger then 2%, we will execute the MPS module again. Once the recursive revision is done, we will execute production performance estimate module to calculate the flow time for each product type and the optimal quantity of WIP to ensure that the MPS is practicable in the shop floor. The case study reveals that compared with the simulation results, the average estimation error of production flow time is about 4.5% when release rule being uniform loading (UL); moreover, the average error is about 2% when adopting CONWIP release rule. In addition, the system throughput rate is about 100% achieved. Hence, the production scheduling system proposed in this thesis can be applied to the environment with demands being periodically changed.