半導體設備昂貴,其設備相關直接成本佔製造成本比例相當高在半導體廠中。因此維持機台設備的高利用率為半導體廠中一重要的課題。預防維護是維持機台設備高利用率一個重要的因素。然而,機台設備預防維護的觸發時間點並不明確,以致於無法安排該時間點資源,影響整條生產線。因此有效預測設備機台預防保養時間點相當重要。 本研究利用灰預測理論進行設備維護時間點預測,並且使用遺傳演算法搜尋灰預測背景值中的適當α值,及利用殘差補償以提升預測準確率。對目前半導體廠三種設備預防維護型態(累積使用片數、累積千瓦數、累積膜厚數)作分析, 結果顯示本研究提出方法較原始的灰預測理論準確。 本研究貢獻為: 提出兩種方式改善灰預測準確度: i. 利用遺傳演算法搜尋適當α值 ii. 利用灰預測對殘差再作一次預測
Equipment costs constitute approximately 3 quarters of overall manufacturing costs in semiconductor manufacturing. Preventive maintenance(PM) is an important cause pertaining to maintain high equipment availability. However PM trigger timing are not clear. Yet, timing is needed for planning of PM resources. Therefore, forecasting equipment preventive maintenance(PM) timing is very important in semiconductor fabrication plant. After having the timing of preventive maintenance, we not only can plan the manpower at the timing to optimize the utility of manpower but also minimize negative impacts on manufacturing efficiency. This study used genetic algorithm to locate optimal parameter, α. and used residual error to compensate original grey model. Tests on three kinds of semiconductor equipment showed significant improvement over the original grey forecast model.