| 摘要: | 光學元件品質的優劣將影響整個系統設計的預期性能。 幾世紀以來幾何光學元件研磨製造的生產技術與設備雖然持續的進步，但是要達到元件設計所要求的各種光學設計和精準度則其生產的基本作業方式與核心技術仍然依照數百年來承傳最基本的工作原理。 幾何光學元件的設計依據系統預期性能的需求對材料、曲率、尺寸與面精度會有所不同。 一件品質良好同時生產穩定的光學元件通常需要由經驗豐富的技師依照前述的設計要求，考慮各種不同光學等級的玻璃或高分子聚合材料之特性﹔調整設定各種設備操控參數組合如研磨設備磨皿與磨仁間的壓力、研磨機擺框之擺輻、偏心連桿偏心量、擺臂長、擺臂配重、細磨鑽石粒的粗糙度和分佈軌跡、拋光作業中氧化鈰溶液的比重、研磨與拋光作業、轉速、和作業溫度等等。然而由技師依經驗嘗試調整的最佳的工作參數組合是在反覆不斷的嚐試下逐漸產生。 光學元件的研磨製造主要包括三大製程，依序為粗磨、精磨﹙細磨﹚及拋光製程，其中精磨製程是決定最終元件成品曲率半徑的重要程序，而且磨後的面粗糙度值也是決定後續拋光作業效率與品質產出的重要因素，故本文探討的主題將由光學鏡面的的精磨製程切入，而實驗設計的目的是利用系統化的工具有效的得到研磨作業的最佳操控參數組合。 論文中實驗最佳操控參數組合的優化過程，主要分為兩階段，第一階段導入Six Sigma管制流程作業方式，歸納出影響光學元件研磨精密製程的主要五項參數設定值﹔依次為研磨時間﹙T﹚、研磨機擺框之擺輻﹙ ﹚、偏心連桿偏心量﹙R﹚、擺臂長﹙ ﹚、擺臂配重﹙W﹚，然後結合Taguchi robust design 實驗計劃法 ﹔進行有規劃的水準調控變化實驗，求取實驗全區域範圍視光學常用凹與凸透鏡球面度數範圍共計57組弧度設計範圍內研磨作業參數應用的最適解。 實驗設計方法選擇若採用一般｢試誤法｣，在五項變數、三階選擇、兩個狀況的情形下可能至少要做﹙53 x 53 =15,625﹚一萬五千次以上的實驗﹔本論文實驗採用Taguchi robust design 的誤差調合式直交實驗計劃法，因此全區域範圍的最適合解在進行﹙18 x 3 =54﹚54次實驗後就便得到適切的結果，而且以S/N值所得到的最適解，因為同時考量了變異數與平均值的影響，因此也是對製程變異不敏感的穩健解。 第二階段的目標，則在求取對光學元件研磨的兩項重要特性目標(表面粗糙度值、曲率半徑誤差值)，可共同達到最佳效果的最適操控參數組合優化。 然而，完美的多目標的最佳操控參數組合並不實際存在，因而最終結果仍繫於參數之間的取捨﹔因此本文應用較為客觀的Artificial Neural Network類神經網路模式來調合實驗結果，捨棄一般的主觀的經驗法則判斷。 最終，實驗結果符合預期成功取得了操控參數設定值之間優化的最佳妥協解。 現今應用科技不斷的快速複製與演進，可預期的生產效率與成本競爭的全球化，光電產業的挑戰將是理論設計與生產實務的結合，提昇產業的競爭力是應並重於反應速度、研發產製能力、穩定的品質、成本的控管等範疇。另外參數優化時間大幅縮短。本論文的目的以系統化的管制流程與實驗參數優化工具，減少人為主觀經驗判斷、增加品質與製程控管能力﹔另外一個應用的範疇為此組生產參數設定可以成為實驗中所使用同型生產機具的出廠標準設定，工程人員可以參考做為在開始作業時的機具參數基本設定，以期設計與生產的整體提昇。 Surfacing technology for optical components has been well established for almost 250 years. The industry has continued to grow vigorously mostly because of new applications. Throughout the long development history, many different designs and materials were applied; however, the process and challenges of today remain similar to those experienced from the very beginning. Processed surface quality is one of the key factors in achieving good optical performance. Although the modern machines are now equipped with better capabilities, optimization of operation conditions by skilled technicians remains a requirement. Especially, the best control parameters require practices that include trade-offs during prototyping and production startup. Moreover, time-consuming trial and error methods based on experience remain a general practice. All optical components follow the same three-phase surfacing process in general including: 1st - Generating, 2nd - Fine grinding, (or smoothing), 3rd - Polishing. The Fine grinding phase was identified as the most critical process for production efficiency, quality yield, and component performance. Thus, this experimental design focuses on the fine grinding process. The experiment design in this paper applied the general use ophthalmic spherical power range as study case. The lenses design including meniscus concave lens for myopia correction and periscopic convex lens for hyperopia or presbyopia correction. There are total 57 sets curvature designs with 0.25D step international ophthalmic standard spherical power range form S-7.00D to S+7.00D. This paper shows how process optimization can be achieved in two steps, the first step using Six Sigma methodology gauges the surfacing process control in order to confirm the five general specified factors that are critical to the surfacing operation. A second effective method coupling the Taguchi experimental design and the most important improvement tools of Six Sigma methodology was then applied. The design plan is based on the use of orthogonal arrays introduced by Taguchi. Through the application of Taguchi’s signal-to-noise (S/N) ratio, we demonstrate that the best parameters design plan from an experimental design can be determined. This has several implications: (1) It reduces the implementation time, (2) it can identify a fractional design that contains the best design plan and that design plan could be studied for full experimentation, (3) within a subset of a fractional design plan, the best design point can be found, and (4) the cost of experimentation is significantly reduced since a minimal number of runs is required to identify the best design point. Finally, this important result helps experimenters to select a fractional design plan that contains the “best design point.” The experiment condition for example, it takes minimum﹙53 x 53 =15,625﹚15,625 experiment trials if using the traditional trial and error methods in order to find the optimal parameters. The fact, the results prove the optimal parameters can be found and confirmed with only (18 x 3 =54) 54 trials according to the design in this paper. The result shows it takes only 0.34% time if the same effect use the traditional trial and error non-specific methods. The traditional control parameters require practices which include trade-offs by skilled senior engineers who are required at this moment to make experiential judgments. In this article, optimized parameters are obtained by applying the mathematical exercise of Non-linear “Artificial Neural Network” to eliminate the subjective judgments. It replaces the errors caused from the experiential judgments made by the expert senior engineers. In terms of the production equipment control and adjustment ability of the newly recruited technician, their capability for exact and reasonable recognition of the production parameters set up is substantially improved. Moreover, the optimal parameters can be applied as the default factory setting in order to be utilized as the reference parameters for general production purposes. |
