模擬輔助類神經產出預測系統於CF廠之應用

A simulation assisted neural-networks estimation system on throughput for TFT-LCD color filter fabs

余正威(Cheng-Wei Yu)

指導教授 ： 黃博滄

中原大學/電機資訊學院/工業與系統工程研究所/碩士(2012年)

https://doi.org/10.6840/cycu201200656

摘要

在許多製造業中，控制投料與在製品(WIP )已經成為一個重要的問題，因投料的變化將會影響在製品的數量，進而改變成本與利潤，為了幫助製造業解決此類問題，學者們紛紛提出各種不同的理論，例如: 投料控制(Order release)、在製品控制(WIP control): CONWIP理論...等等，而大多研究是將在製品控制與投料控制兩個理論分別探討，並無研究將在製品控制和投料控制兩個理論結合探討之，這些理論必須要經過風險評估與分析後，才可實際運用。
於近年來研究中，模擬已成為工廠內做為決策分析的重要工具，然而，模擬工具需要去建立、驗證與實驗，這些步驟需要耗費時間，無法快速的得知所需要的訊息，如何去省下模擬所需要的時間，將成為一門重要課題。因此，本研究將利用AutoMod模擬軟體並結合人工類神經網路(Artificial Neural Network, ANN)，建立一套預測系統(SANNF)，透過SANNF預測系統使管理者能迅速的找到更佳的決策方案，並利用此系統將投料控制與在製品控制做結合，找出其平衡點，開發出權衡系統(Trad-off system)。此系統能使決策者在不用承擔風險之下，更快的進行決策分析(What-if analysis)，亦可以找出投料控制與在製品控制之平衡點，相信本研究能使工廠達到更高的效益。

關鍵字

彩色濾光片 ； 類神經網路 ； 模擬 ； 投料控制 ； CONWIP

並列摘要

In the manufacturing industry, scholars have proposed various theories to try and solve fab problems; among theories proposed, WIP control and order release are two of the more important ones. These theories must evaluate performance to reduce the risk of failure，and can be used in real fabs.
However, simulation has also been widely discussed to evaluate performance in recent years. Since simulation software AutoMod models always take time to run simulations, they cannot quickly provide information to planners to adjust the parameters. In order to solve this problem, this study proposes an algorithm for the Neural Network (NN) theory. The integration of ANN and simulation was conducted to build a forecasting system. Further, there are trade-offs between cost and effect for WIP and throughput. Management may need to stimulate environments to find the existing trade-off points and determine the actions needed to minimize their effect on WIP and throughput. Thus, the SANNF system is integrated with Artificial Neural Network and AutoMod Software. The combination of the SANNF system and mathematical theory resulted in a new trade-off system.
This system enables planners to conduct what-if analysis for the production control policies without disturbing operations in the real plant. Better alternatives leading to higher system throughput can be identified by using trade-off system in the fab.

並列關鍵字

Neural Network ； Color Filter ； CONWIP ； Order release ； Simulation

參考文獻

Burman, D. Y., Gurrola, F. J., Norzari, G. A., Sathaye, S., 1986. Performance Analysis techniques for IC manufacturing lines. AT&T Technical Journal, 65(4), 46-57.

Cao, D., Chen, M., 2005. A mixed integer programming model for a two line CONWIP-based production and assembly strategy. International Journal of Production Economics, 95(3), 3317-326.

Chen, J. C., Chen, C. C., Mo, C. N., Peng, T. W., Chao, G. C., Sun, C. J., Chao, K. J., Chien, J. W., Wei, Y. Y., Feng, C. H., Liu, Y. C., Chiang, M. T., 2009. Effective order release and dispatching for color filter fabs. Society for Information Display International Symposium (SID), San Antonio, Texas, USA.

Chen, J. C., Chen, C. W., 2010. Capacity planning of serial and batch machines with capability constraints for wafer fabrication plants, International Journal of Production Research, 48(11), 3207-3223.

Chen, J. C., Su, L. H., Chen, C. C., Peng, T. W., Chao, G. C., 2011. Release planning for color filter fabs based on iterative simulation, Journal of the Society for Information Display, 19(7), 480-487.