視覺化分析輔助PCB新製程研發之研究

Using visualization to support development of PCB new manufacturing process

王俊程(Jun-Chen Wang)

指導教授 ： 孫天龍

元智大學/工程學院/工業工程與管理學系/碩士(2004年)

摘要

近年來台灣的印刷電路板(Printed Circuit Board, PCB)產業已成為技術成熟的產業，在產品生命週期日漸縮短、同業競爭激烈的環境下，如何快速研發生產成本更低，品質良率更好之新製程，便成為各廠商提昇競爭優勢且獲利的關鍵。在PCB新製程研發流程中，研發人員要找出影響品質特性之製程控制因子時，常常會忽略了一些「隱性」之因子，這些因子看起來不重要，但因交互作用反而變成影響製程的重要因子。利用分析新製程實驗批次資料所獲得之客觀資訊，可輔助研發人員找出這些隱性的製程因子，但由於成本的考量，新製程研發時多半僅能進行少數幾次的實驗批次，因此資料筆數不多，導致一般常用來分析生產製程資料的資料探勘法不適用於分析新製程實驗批次資料。視覺化資料分析技術可將資料以圖形化方式呈現，並利用人類對於圖形中之視覺特徵快速辨識能力，從圖形中了解資料內隱含之資訊，且可藉著分析過程中使用者與電腦互動達到更彈性、更具創意的資料分析，本研究因此發展一資料視覺方法輔助分析PCB新製程之實驗批次資料。我們將PCB新製程實驗批資料以非等距平行座標圖呈現，並提出平行座標圖上之折線相似度比對方法，接著我們討論三種視覺分析方法，第一種方法利用圖形觀察相關係數無法呈現的資料相關細節。第二種方法利用研發人員的視覺辨識能力以及創造性思考能力定義出平行座標圖上的參考比對折線，再利用電腦找出與參考折線形狀相似之折線，以探討具交互作用之隱性製程因子。第三種方法利用圖形觀察製程因子間的相關性，配合研發人員之製程知識，討論因子間的交互作用。

關鍵字

印刷電路板 ； 視覺化資料分析 ； 平行座標圖

並列摘要

Developing a new manufacturing process with lower cost and higher quality is critical for PCB manufacturing companies to promote competitive in the market. One challenging task in developing a new process is to find out the controlling factors that will affect the quality features of the manufacturing process. Although the process engineers could pick up most of the important factors by experiences, they often miss some invisible factors. Those invisible factors may not seem important at first glance, yet the interaction among them will become a major factor contributing to the quality problem. The data collected from experimental batches in developing a new manufacturing process may contain useful information for researchers to figure out the invisible factors. Due to cost concern, however, only a limited number of experimental batches could be conducted in developing a new manufacturing process. As such, the number of data is not enough for data mining methods. Data visualization converts the numerical data into graphs to take advantage of human’s natural strength in visual pattern recognition and creative thinking. This research presents a visualization-based method for analyzing manufacturing data collected from a new PCB manufacturing process. A non-equal-distance parallel coordinate is proposed to display the PCB manufacturing data. Also a method to compare the similarity of poly-lines on the parallel coordinate chart is developed. To find out manufacturing factors that may interact with a query factor, the process engineers will define a reference poly-line on the parallel coordinate according to their experiences about interactions. The computer then searches the manufacturing data to find out those control factors whose poly-lines on the parallel coordinate chart are similar to the reference line. These poly-lines are then displayed to let process engineers further examine the interactions between the query factor and the founded factors.

並列關鍵字

PCB ； data visualization ； parallel coordinates

參考文獻

蘇星樺，2003，「視覺化資料探勘輔助偵測PCB新製程之不良成因」，碩士論文，元智大學工業工程與工程管理研究所。

Braha, D. and A. Shmilovici，“Data mining for improving a cleaning process in the semiconductor industry,“ IEEE Transactions on Semiconductor Manufacturing, 15, February, 91-101, (2002).

Gardner, M. and J. Bieker, “Data mining solves tough semiconductor manufacturing problems, “proceedings of the IEEE International Synposiom on Semiconductor Manufacturing Conference, 2000, 376-383.

Goel, A., C. Baker, C.A. Shaffer, B. Grossman, B., Haftka, W.H. Mason, and L.T. Watson, “VizCraft: A Multidimensional visualization tool for aircraft configuration design,” Visualization ''99. Proceedings, 1999, 425-555.

Inselberg, A., “ The Plane with Parallel Coordinates, “ The Visual Computer, 1, 69-91, (1985).