應用資料探勘技術探討扣件關鍵檢驗參數

Using Data Mining Techniques to Explore the Critical Inspection Parameters in Fastener Manufacturing.

莊敏姿(Min-Tzu Chuang)

指導教授 ： 劉建浩

國立臺北科技大學/管理學院/工業工程與管理系碩士班/碩士(2014年)

摘要

台灣扣件的生產量佔全世界超過六分之一，年產值更高達千億元新台幣。但是近年來受到中國大陸及東南亞國家的雙重夾擊，而我國的廠商大都是代工廠，勢必要進行轉型，朝向高附加價值的產品進行研發與生產。然而，其生產效率有遲緩現象，究其原因，主要因為達到高品質的自我要求，相關製程的檢驗極為嚴格，許多製程都因檢驗程序未完成而無法順利展開。
過去關鍵製程參數的決定多仰賴工程師的經驗，但隨著新製程不斷的開發，此種模式已不符企業要求。資料探勘為一種將資料轉化為知識的方法，本研究將採用其支配性約略集合 (Dominance Rough Set Analysis, DRSA)模式，此模式的優點是可以處理資料間不一致的問題。利用支配性約略集合找出關鍵的檢驗參數，提供產品可允收標準，減少不必要的製程檢驗程序，同時建議關鍵製程參數設置的重點。並且以倒傳遞類神經網路驗證其分類正確率。本論文以某扣件工廠為例，建構尋找關鍵檢驗參數系統，結果顯示鈑金到頂端的距離、大孔同心度、耳下安全島、尾部外徑、耳上平面、花齒外徑、身體長度、花齒長度、頭部厚度、安裝負載經過訓練後分類正確率高達97.6%，證明此方法可以得更好的分類結果。

關鍵字

扣件 ； 關鍵參數 ； 支配性約略集合 ； 倒傳遞類神經網路

並列摘要

Taiwan has produced more than one sixth of the fastener in the world. The revenue exceeded NT$100 billion in recent years. Recently, the development of Mainland China and Southeast countries have severely threaded the survival of Taiwan companies and most of the companies are suppliers and do not have their own brands. In order to keep the advantages of Taiwan fastener industry, the companies have to transform into higher levels of products and devoted to research and development. However, firms have spent most of their efforts on marketing expanding and new product development, thus their production efficiency has slow down. After carefully examination, they found the root cause was due to many inspections delayed the production rate. Therefore, how to identify the key inspection parameters and reduce some unnecessary inspection procedures are the objectives of this study.
The key manufacturing parameters were decided by engineer experience in the past, but this method is not fit in today’s enterprise requirements. Data mining is a technique that transfers data into knowledge. Through data collection and analysis, analyzers can find the association rules between the data. This study will apply Dominance Rough Set Analysis (DRSA) to decide the key inspection parameters and provide acceptable standard. DRSA is a derivation of Classical Rough Set Analysis (CRSA) but DRSA can handle the inconsistent problems. The contribution is that the case company can obtain the critical inspection parameters, decrease the unnecessary examinations, and increase its production rate. At the same time, suggesting what manufacturing process focus. To verify the results, this study uses back-propagation neural network training model to get correct rate. Finally, this proposal compares the correct rate of CRSA with DRSA results. In the study, a fastener company is chosen as an example. According to this real data, the proposed approach provides better classification results.

並列關鍵字

Fasteners ； Critical Inspection Parameters ； Dominance Rough Set Analysis ； Back-Propagation Neural Network

參考文獻

38.Hassanien, A.E., Abdelhafez, M.E., Own, H.S. (2008), “Rough sets data analysis in knowledge discovery: a case of Kuwaiti diabetic children patients”, Advances in Fuzzy Systems, 8(2): 1-13.

5.林恆勝、許源泉、楊東昇、楊政峰、歐聖堂 (2007)，「六角螺栓近淨形頂鍛成形模擬」，鍛造， 第16卷，第1期， 第5-16頁。

3.李榮顯 、陳彥儒、王水鐸 (2004)，「利用凹槽設計於圓柱鍛粗試驗之可行性探討」， 鍛造，第13卷，第4期，第29-34頁。

8.郭展銓 (2011)，「失效模式效應分析在扣件熱處理製程上之應用」，品質月刊，第47卷，第12期，第23-28頁。

7.陳政順、沈忠信 (2009)，「電腦連接器卡扣結構件之分析與最佳化設計」，技術學刊，第24卷，第2期， 第99-105頁。