- 學位論文
類神經網路應用於光碟旋塗之配方最佳化搜尋
Optimal Recipe Search for Spin Coating Process of Disc Manufacture Using Neural Networks
摘要
旋轉塗佈被廣泛應用於國內各大科技廠，國內先進光碟廠將旋轉塗佈應用於染料塗佈以及貼合膠塗佈，主要是因為其具有高均勻性以及低成本兩大優點，工程師僅需針對塗佈參數(加速時間、停留時間以及旋轉轉速等)做適當調整，即可完成光碟片之塗佈工程。而要如何將上述參數做最佳化搭配，仍需靠工程師利用試誤法方式做配方搜尋。試誤法雖然可以搜尋出最佳化配方搭配，但其仍得仰賴經驗法則。有經驗的工程師可以很快的搜尋出最佳配方，但對於較資淺的工程師而言，卻需要耗費更多的時間以及成本去達到配方搜尋之目的，故此法並不符合經濟效應。 本論文所提出的方法，是先將一系列有規劃之實驗數據運用徑向基底網路(RBFN)建模，在智慧型基因演算法(TGA)演化出最佳化參數後，代入此製程模型中做驗證。如能符合所設定之條件，即完成最佳化配方搜尋。經由實驗驗證，本論文所提出之實驗方法，不僅可有效地搜尋出最佳化配方，目標準確度也符合作業所設定之誤差範圍(±0.003)，與試誤法作比較，所得之結果亦十分接近，本論文所提之方法可大幅縮短配方搜尋時間，在生產作業目標值改變時，工程師可運用少量的實驗規劃，配合本論文所提出之方法，可達到最佳化配方搜尋目的。
並列摘要
Spin coating has been widely used in big domestic factories such as the advanced optical-disc manufacturers. The manufacturers apply spin coating to dye coating and bonding coating for two main advantages, namely the high uniformity and the low cost. With spin coating, engineers will only need to properly adjust the coating parameters, acceleration time (ACC Time), step time and spin speed to complete the discs’ coating process. Nevertheless, optimizing the mentioned parameters is involved with the use of the trial-and-error method by the engineers during the recipe search. Although the trial-and-error method might search out the optimal recipe, it will, however, heavily rely on empirical rules. To an experienced engineer, he might be able to quickly search out an optimal recipe, but to a less experienced engineer, it will take him more time and cost to reach the goal of recipe search. Hence, the trial-and-error method usually does not render economical efficiency. In this thesis, we first propose applying a radial basis function network (RBFN) to a series of preplanned experimental data, in establishing a model, and adopting an intelligent-type Taguchi genetic algorithm (TGA) to evolve the optimal parameters. Next, we substitute the parameters into the process model for verification. If the model conforms to the set conditions, the optimal recipe search is then completed. Through our experimental verification, we observe that the proposed method not only can effectively search out the optimal recipe but also renders a goal conforming to the discrepancy range of (±0.003) set by the process. In comparison, the results obtained from the proposed method come quite close to that of the trial-and-error method; furthermore, the proposed method can greatly reduce the recipe search time. When the production process target alters, engineers equipped with the proposed method, unlike with the other method, will be able to reach the optimal recipe search goal using only a small amount of experimental planning.
參考文獻
延伸閱讀
- 林永威（2010）。類神經網路應用於透明導電膜連續濺鍍法之配方最佳化搜尋〔碩士論文，中原大學〕。華藝線上圖書館。https://doi.org/10.6840/cycu201000960
- Liu, L. C. (2019). 開發類神經網路非線性分類器及降低其複雜度之方法並應用於高速光傳輸系統 [master's thesis, National Chiao Tung University]. Airiti Library. https://www.airitilibrary.com/Article/Detail?DocID=U0030-0306202016341678
- 李宜峻（2016）。利用類神經網路於記憶體匯流排之訊號完整度分析〔碩士論文，國立臺灣大學〕。華藝線上圖書館。https://doi.org/10.6342/NTU201603654
- 王亭幼（2019）。運用多元卷積類神經網路架構於多光譜遙測影像之物質分類〔碩士論文，國立臺中科技大學〕。華藝線上圖書館。https://www.airitilibrary.com/Article/Detail?DocID=U0061-1408201913360300
- 王仁達（1997）。A Neural Network Approach for Detecting Changes in Process Mean : the Investigation of Design Strategy〔碩士論文，元智大學〕。華藝線上圖書館。https://www.airitilibrary.com/Article/Detail?DocID=U0009-0112200611340153