|Abstract: ||現今電子產品逐漸走向輕薄設計，其基材逐漸轉向塑膠等軟性材料。然而市面上製作軟性電子大多以黃光蝕刻製程，其步驟繁複且成本高，故捲對捲 (roll-to-roll, R2R) 印刷為軟性電子未來可能量產的技術之一。而大多軟性電子材料為多層結構，故在印刷控制上必須做到好的對位控制 (Registration control) 以降低對位誤差。然而在捲對捲多層印刷機台架設影像感測器會遇到兩個挑戰─第一，印刷第一層材料時同時印刷對位記號，進一步利用影像處理找出圖型重心以取得參考位置。然而印刷過程中若對位記號有缺陷則影像特徵會出現誤差；第二，影像感測器在FOV (Field of view) 為 2.5*2.5 ?mm?^2 的情況下，對於傳輸方向最大速度 v_web=10 m/min 其取樣時間必須小於 10 ms/frame 內。本研究目的為研發一個適用於捲對捲印刷位置影像感測系統；藉由對於拍攝之對位記號進行影像處理統計分析，設計最佳化對位記號，以降低圖案誤差造成之感測誤差。另外藉由即時作業系統 (Real-time Operating System) 及平行運算影像處理方法使得從拍照、影像處理及下控制指令之總時間能落在 10 ms/frame 以下。在對位記號圖型設計的結果上，得知當直線資料點數 n 越多時其特徵估測值 (θ ?,ρ ? ) 之變異量皆會越小，其值越接近實際值。在影像感測性能方面，其影像處理總執行時間約落在5.6 ms，而最差的時間約為7.7 ms，故此感測系統之取樣時間符合我們的目標 8 ms/frame 以下 (假設下控制指令為 2 ms 以下)。|
Electronic products are moving toward thin and light design at present, and its substrates gradually change to flexible materials such as plastics. Most of the flexible electronics produced on the market are made by lithography or etching process, which is complicated and expensive, therefore roll-to-roll (R2R) printing is one of the technologies to produce the flexible electronics in large quantities in the future. Also due to multi-layer structure of flexible electronics, it must be performed well for registration control to reduce alignment error on the printing control.However, there are two challenge for R2R multi-layer printing with image visual sensor – firstly, as the first layer of material is printed, the alignment marker is coated at the same time, and then the reference position will be obtained through using image processing to find the centroid of the alignment marker. But the image features will get error when the alignment marker coated in the R2R printing process is defective. Secondly, in the case of 2.5*2.5 ?mm?^2 of FOV (Field of view) for the image sensor, the sampling time must be less than 10 ms/frame for the maximum speed v_web=10 m/min in the transmission direction.The purpose of this study is to develop an image sensing system suitable for R2R printing position; the optimized alignment marker is designed by statistical analysis of the image processing to reduce the sensing error caused by the pattern error. In addition, the total executing time from image capturing, processing, to ordering the control commands can be reduced to less than 10 ms/frame by the real-time operating system and the parallel computing image processing method.On the result of the design of the alignment mark, it is shown that the more the number of linear data points n is, the smaller the variation of the feature estimated value will be, and its value will be closer to the actual value. In terms of image sensing performance, the total executing time of image processing is about 5.6 ms, and the worst time is about 7.7 ms. Therefore, the sampling time of the sensing system meet our target of 8 ms/frame (assuming ordering the control command is less than 2 ms).