採迴歸樹進行規則探勘以有效同時降低多種紡織瑕疵;Adopting regression tree for rules mining to effectively reduce various fabric defects simultaneously

NCU Institutional Repository > 資訊電機學院 > 資訊工程研究所 > 博碩士論文 > Item 987654321/89980

請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/89980

題名:	採迴歸樹進行規則探勘以有效同時降低多種紡織瑕疵;Adopting regression tree for rules mining to effectively reduce various fabric defects simultaneously
作者:	楊華升;Yang, Hua-Sheng
貢獻者:	資訊工程學系
關鍵詞:	參數組合;規則探勘;特徵選取;決策樹;parameter combination;rule mining;feature selection;decision tree
日期:	2022-08-22
上傳時間:	2022-10-04 12:06:26 (UTC+8)
出版者:	國立中央大學
摘要:	紡織業為臺灣不可或缺的產業，如今生產製程遭遇快速反應、品質穩定、交期掌控的三大關鍵挑戰，在機臺的參數設定階段找出瑕疵率較低的最佳參數範圍組合是維持品質穩定的策略之一。規則探勘為本研究鎖定重點，資料庫中瑕疵總數排名前四的瑕疵種類，針對其相關織物性質與機器參數進行數據分析，研究目的在於找出對各類瑕疵值最具影響力的重要特徵集，並將得出的特徵整併成範圍規則型式，最終各瑕疵種類規則需依照特定方式進行合併，旨在有效同時降低四種瑕疵並帶來全域效益，效益意即相較原始無設定此規則時減少多少總瑕疵比例。實驗綜合多樣分析，包含採取不同特徵選取的[criterion, scoring](以中括號代表特徵選取所需的參數對，其餘處亦採取此表達格式)、制定自決策樹提煉候選規則的機制、擴大或縮小候選規則的參數範圍、合併規則的組合方式等。採用嵌套交叉驗證後，最終在各時間拆分點皆可探勘出效益大於等於0.8，或者支持度大於等於0.5或近似0.5的全域規則。支持度用來評估規則涵蓋測試資料的多寡，亦可解釋為規則對機器狀態是否穩定的影響程度。 ;The textile industry is an indispensable industry in Taiwan. Today, the production process encounters three key challenges: rapid response, stable quality, and delivery control. Finding the best parameter range combination with a lower defect rate in the parameter setting stage of the machine is the key to maintaining stable quality. Rule mining is the focus of this research. The total number of defects in the database ranks in the top four types of defects, and data analysis is carried out according to their related fabric properties and machine parameters. The purpose of the research is to find out the most influential feature set for each type of defect value, and integrate the obtained features into a rule type. In the end, the rules for each type of flaws need to be combined in a specific way, aiming to effectively reduce the four flaws at the same time and bring global benefits. How much to reduce the percentage of total imperfections. The experiment comprehensively analyzes various aspects, including the selection of different features [criterion, scoring], the development of a self-decision tree mechanism to extract candidate rules, the expansion or reduction of the parameter range of candidate rules, and the combination of merging rules, etc. After using nested cross-validation, the global rules with benefit greater than or equal to 0.8, or support greater than or equal to 0.5 or approximately 0.5 can be mined at each time split point. The support degree is used to evaluate the amount of test data covered by the rule, and it can also be interpreted as the degree of influence of the rule on the stability of the machine state.
顯示於類別:	[資訊工程研究所] 博碩士論文

文件中的檔案:

檔案	描述	大小	格式	瀏覽次數
index.html		0Kb	HTML	32	檢視/開啟

在NCUIR中所有的資料項目都受到原著作權保護.

社群 sharing

資料載入中.....