隱藏式馬可夫模型用於印刷電路板廠在製品圖樣視覺化之辨識

PCB WIP pattern identification and visualization using Hidden Markov Model

陳志蒼(Chin-Tsang Chen)

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(2001年)

摘要

資料視覺化係將大量資料轉換成圖表，利用人類與生俱來的快速視覺識別能力，更容易、更有效率的判讀資料，在之前的研究中【宋保全,1999】已完成印刷電路板(PCB)廠在製狀況資料（WIP）之視覺化探勘環境，讓生管人員快速瞭解生產狀況，掌握生產資訊，發掘潛在問題。在製品圖樣是將產品的在製品狀況資料由數據化的資料轉化成一點點由灰階代表程度的方格，此種小圖稱為對應表格。依據眾微圖設計【Tuffe,1983】的概念，可將每個產品的對應表格呈現於在一個螢幕上，以利生管人員解讀資料。
由於圖樣的形狀是依據生產情形所決定，故具有類似圖樣的產品，其生產過程應該有一定的相似度。利用此特性，本計劃第二年之工作重點係利用隱藏式馬可夫模型（Hidden Markov Model）發展一自動WIP圖形模式辨識系統，將WIP表格依其會呈現之特徵自動分組並將同一類型之WIP表格圖形排列在一起，以增進使用者視覺效率。

關鍵字

資料視覺 ； 隱藏式馬可夫模型

並列摘要

Data visualization converts large amount of data into graphs to take advantages of human''s natural strength in rapid visual pattern recognition to enable easier, effective comprehension of data. In a previous research【sung ,1999】,we have developed a PCB WIP pattern visualization system that provides production managers a means to rapidly understand current production status, discover useful production information, and find out possible production problems from large amount of production data. The WIP Pattern of a product is converted into a small graph called a “smashed table” by smashing numbers in a table into small square dots. The smashed tables of multiple products are on the placed on the display screen according to the small multiple design principle【Tuffe,1983】.
Since WIP patterns are closely related to production status, products with similar WIP patterns should have similar production status. Based on this property, in this research we use the Hidden Markov Model to develop an automatic WIP pattern recognition technique to identify products that have similar WIP patterns and group these products together to enhance comparison and improve the effectiveness of visualization.

並列關鍵字

HASH(0xbeddc64)

參考文獻

宋保全，「以視覺化資料探勘輔助印刷電路板廠之生產管理」，私立元智大學，碩士論文，民國89年。

郭玟琳，「以資料視覺輔能規劃的決策」，私立元智大學，碩士論文，民國88年。

Andrews D. F., “Plots of High-Dimensional Data,” Biometrics, 29, pp. 133-139, 1972.

Baum L. E. and J. A. Egon, “An intequality with applications to statistical estimation for probabilistic functions of a Markov process and to a model for ecology, ” Bull. Amer. Meteorol. Soc., vol.73, pp.360-363, 1967.

Beddow J. , “Shape coding of mulyidimensional Data on a Mircocomputer Display,” Visualization ’90, San Francisco, CA, 1990, pp.238-246.