近年來電子產品之元件為因應下游產品輕、薄、短、小的趨勢，已逐漸由晶片型元件取代了傳統型元件。本研究的對陶瓷電容器，也由過去的單一重疊面積的圖板型電容器，縮小成為現在的重覆堆疊式的積層陶瓷電容器。此積層陶瓷電容器目前業界大多採用溼式印刷淋膜製程從事生產，在此製程中電極印刷作業於全自動化的連續印制過程中完成，其製成程序是在膜板上淋膜一固定厚度的瓷漿，而後印刷一層銀鈀電極再淋膜上一層瓷漿，依序反覆地來回製作，直到所需的設計層數為止。此種作業將造成每層印刷後的電極圖形會被下一層瓷漿所遮蓋掉，而無法確定每層的印刷電極的位置是否固定不變，因而產生無法對此印電極之位置之位置是否發生異常偏移的情況進行製程管制的問題。 本研究運用電腦視覺技術與統計製程管制方涇來解決上述製程所面臨的問題，期能有效地偵測與管制被遮蓋的銀鈀電極產生過度偏移的情形。本研究的方法分成三個階段進行，第一階段為解決承載印刷電極之模板的定位問題，以旋轉與平移矩陣進行影像位置補償，來固定不同時段所進行印刷作業之模板的位置；第二階段為找出模板上各印刷電極圖案之位置，以特徵分析法進行各層印刷電極圖案的幾何中心（重心）之位置量測；第三階段為以統計製程管制來偵測製程中的印刷電極作業是否已產生異常偏移，利用具偵測微量偏移特性之累和（Cumulative-Sum）管制圖監督各層印刷電極圖案是否已產生過度偏移的情況。經實驗結果分析顯示，本研究的方法能有效地偵測與管制印刷電極作業產生60μm以上異常偏移的製程，此研究結果不僅能解決上述問題，而且可以提供給類似具有被遮蓋特性的多層堆疊式生產製程做為製程管制與檢驗作業的參考。

　　　　　Since recently developed electronic products are extremely thin and light, the chip type components have replaced the traditional components. In this research, the observations are ceramic capacitors which have shrank from single-layer disk capacitors to multi-layer ceramic capacitors. In the multi-layer stack production process of multi-layer ceramic capacitor, silver-palladium electrodes and ceramic pastes are stacked respectively. Since the printed electrode is covered by ceramic paste, it makes it difficult to make sure that the electrode of the upper layer is located at the right place as that of the lower layer. Not being able to inspect whether the locations of the printed electrodes at different layers have shifted over the desired level causes a serious inspection problem. Different shift amounts of the silver-palladium electrodes may result in different degrees of defects. If the electrodes shift slightly (within 10-50μm), the capacity of multi-layer ceramic capacitors will decrease. If the shift amounts are medium (between 50μm and the marginal of chip), the insulation resistance (IR) value of the chip will decrease and the increasing temperature of the board of the capacitors installed may burn out the board. If the electrodes shift seriously (more than the marginal of chip), it will cause the circuit short and the charge and storage functions out of order after electroplate operations. To improve the above defects, this research applies two computer vision techniques to solve the over-shifted problem of the printed electrodes. The template match and feature analysis algorithms use the edge of die as the base line to detect the shift amounts, their performances are evaluated in the research. Finally, the cumulative sum (CUSUM) control chart with fast initial response (FIR) has demonstrated great performance to monitor the over-shifted problem of the printed electrode process. The need for compact, light, thin, and high capacity electronic equipment and consumer electronic products has greatly increased the importance of the multi-layer stack production process. This research contributes a solution to a common inspection problem of the process.