- 學位論文
應用適應性粒子群最佳化於真圓度量測
Applying Adaptive PSO on Roundness Measurement
摘要
曲線擬合（Curve fitting）係指一組數據，找出一組資料集合（Dataset）為最逼近之相似解，本論文以機器視覺技術攫取三吋矽晶圓（Silicon wafer）之邊緣點座標，利用空間轉換成最大內接圓函式為研究對象，以粒子群最佳化演算法及其改良於此非線性極大化問題，同時比較不同提出方法之良窳。由於矽晶圓切片後會出現些許製程變異，導致細微邊緣瑕疵與橢圓現象，而其不良品可由重新外徑研磨加以回收再利用（Reclaim wafer）。本研究藉由晶圓影像攫取和資料前處理，用傳統粒子群最佳化更新法則為基礎，提出藉由分群方式取得動態刪除粒子數以收加速演算法收斂之效，並建議使用者該問題所需之最佳化參數（粒子數），由於須針對問題搭配參數組合之最佳化演算法均有高度仰賴實驗設計劣勢，在本研究中有超過一種面向之改良且具備免實驗設計之適應性（Adaptability）功能，並於最大內接圓模型為研究對象（目標函式）下獲得比較，實證中快速取得的最佳化參數可供支持決策，或結合模式搜尋之全域學習法則（HJ-PSO）彌補原始區域搜尋之不足，而適用不完整資料的灰關聯分析也在本研究中加入，其廣義性（Generalization）能迎合調整後之不完整母體維度與分群機制，相較原始粒子群最佳化演算法，符合快速初步了解一未知問題，且免於考量轉換資料成原求解空間的預期效果。
並列摘要
Inspection on silicon wafers is a complex and important process for semiconductor manufacturers. Optimally manufacturing each wafer to overcome the quartz shortages is tantamount to achieve maximum total profit in practice. Roundness, particularly the roundness of silicon wafers remaining a bottleneck for reclaiming wafer, is a very costly and crucial step for increasing yield. In particular, inspecting post-slicing process of wafers can be considered as a non-linear problem with a specified roundness measure. Therefore, this study proposes heuristic and adaptive methods that rapidly converge with high accuracy and low cost. The proposed methods incorporate the Hooke-Jeeves pattern search with Particle Swarm Optimization in comparison of convergent performance. A substantial amount of effort has been expended to alleviate the redundancy than the former [18] involved. This study primarily focuses on mixture algorithms for measuring roundness of silicon wafers and competes the performance with accuracy (efficiency) through visual inspection. A set of experiments is conducted to verify the feasibility under varied schemes. Definitively, experimental results reveal that the proposed method is superior in terms of execution time and solution quality.