請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/72411
標題: | 具有多對多資料特性之紡織製程瑕疵肇因分析 Root Cause Analysis On Weaving Processes with Many-to-many Data Structure |
作者: | Shao-Chi Hung 洪紹綺 |
指導教授: | 洪一薰 |
關鍵字: | 梭織製程分析,產量損失分析,決策樹,多對多資料, Weaving process analysis,Yield loss analysis,Decision tree,Many-to-many data, |
出版年 : | 2018 |
學位: | 碩士 |
摘要: | 紡織產業梭織製程包括以下四個主要工段:整經、漿紗、併經以及織造,傳統業界現行做法為:將符合訂單規格的原料,依序經歷四個加工工段後,完成完整可供出貨的布疋,再以整塊布為單位進行品質檢驗,此時若發現有瑕疵,只能裁剪布疋,部份報廢或另外加工處理,對工廠無疑是額外的成本負擔。在近幾年快時尚產業衝擊下,紡織業生產週期巨幅縮減,出貨速度與品質須同時兼顧,如何在實際生產時減少整塊布的瑕疵率,變得極為重要,然而傳統統計方法是透過不同自變數(Independent Variables)造成應變數(Dependent Variables)的變化關係建立品質模型,在紡織織造前的整經、漿紗與併經,會將許多來自不同生產參數的小軸,合併在一起,變成一個大軸,而來自相同生產參數的一個大軸,會再織造成一塊帶有數種不同瑕疵類型的布,如何考量多對多資料紀錄 (Many-to-many data structure) 的生產特性下,分析生產參數與品質的關聯性,是一大挑戰。本研究先以布為單位定義瑕疵率,再以敘述統計(Descriptive Statistics)的方式對被併到相同大軸的小軸進行資料前處理,之後透過統計機器學習方法分析,以決策樹(Decision Tree)挑選各工段中,影響品質的重要製程參數與關鍵機台組合,並建立其與織品品質的統計關聯模型,進一步提供重要因子的最佳參數條件,降低瑕疵率,提高生產穩定性。 Weaving manufacturing process contains four sequential steps, namely warping, sizing, beaming, and weaving. Products are then inspected, and those with flaws are scrapped. Since the yield loss is costly, it is of great interest to identify equipment-related and parameter-related causes critical to product defects in the process of manufacturing. However, in reality, firms often neglect the importance of defect causes analysis. Quality control is now based on human experiences of operators, making it difficult to be regarded as Standard Operating Procedures (SOP). Moreover, not only the Work-in-Process (WIP) between weaving process but also the flaw records on one textile has the characteristic of many-to-many data structure which would lead a confusion to quality analysis. In this study, we firstly redefine the yield. Next we use Descriptive Statistics to formulate the relation of process parameters in combined spools. Finally we identify the important parameter-related and machine-related causes and then propose the statistical model to link the important process parameters with the product quality by Decision Tree, widely used and effective statistical machine learning technology. Finally, we suggest a set of production parameters based on the proposed model to reduce the defect rate and enhance the stability of the manufacturing process. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/72411 |
DOI: | 10.6342/NTU201803226 |
全文授權: | 有償授權 |
顯示於系所單位: | 工業工程學研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-107-1.pdf 目前未授權公開取用 | 6.53 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。