利用需求資料提高限制理論需求拉動補貨策略之緩衝管理的效能

Abstract

限制理論所建議的需求拉動補貨模式配合緩衝管理的方法，由於觀念簡單且易於 執行，已在不同的企業中試行而得到很好的結果。但傳統的緩衝管理方法僅依目前的存 貨水位決定是否調整目標庫存，在將之應用於補貨前置時間長、需求變異大且生命週期 短的半導體產品時，可能因未考慮需求的變化情形而導致目標庫存的調整不及，造成缺 貨或存貨積壓過多的情況；若能提早知道需求變化的趨勢，則可以及早因應。 需求的變化情形可由已得的歷史資料推測或是由市場的需求預測而得知，因此本 計畫預計利用兩年的時間提出兩種修正緩衝管理的方法，分別在是否能夠取得市場需求 預測的情況下改善需求拉動補貨模式中的緩衝管理方法，以提此補貨模式的運作效率。 在第一年的計畫中，本計畫針對無法取得市場的需求預測的情況，採用指數加權移動平 均監控需求的趨勢，依據需求的變化情形與目前的存貨水準，決定調整目標庫存的時機。 而在第二年的計畫中，本計畫更進一步針對顧客能夠提供具有一定可信程度的市場需求 預測資訊的情況，設計一套方法將市場的需求預測資訊加入於第一年所設計的緩衝管理 改良方法中，以同時決定目標庫存的調整時機與調整幅度，使得需求拉動補貨模式能運 作得更有效率。本計畫將以實際取自業界的產品需求資料驗證本計畫所提出之方法的有 效性與可行性，同時也將採用模擬的數據探討所提出之方法的適用情境。

Theory of Constraints (TOC) suggests to using demand-pull approach combined with buffer management to effectively manage inventory. Due to its effectiveness and simplicity, this approach has been successfully applied to many companies in practice. However, in the traditional buffer management mechanism, the target inventory level (buffer size) is adjusted based on current inventory penetration level. When applying it to products with long replenishment lead time, volatile demand and short product life cycle, this buffer adjustment mechanism may not be able to catch the demand variability in time thus may lead to situations with shortage or with excess stock on-hand. The demand trend may be estimated by historical data and/or by forecasts. This study plans to design two approaches to increase the effectiveness of demand-pull approach by improving the buffer management mechanism. In the first year, this study proposed to use exponential weighted moving average (EWMA) to monitor the demand trend for this situation when only historical demand data is available. In this method, the timing for buffer adjustment depends on both the demand trend revealed by EWMA and the inventory penetration level so that the buffer can be adjusted in time to react exist increasing or decreasing trends shown in demand. In the second year, this study plans to extend the EWMA approach proposed in the first year by including the market demand forecasts shared by downstream customers. In this method, both the timing to adjust buffer and the amount of buffer to be adjusted can be determined by fully utilizing the available information. This study will use real data retrieved from some wafer foundry in Taiwan as well as the simulated data to demonstrate the effectiveness of the proposed approaches.