臺灣紡織品是世界機能性紡織品消費市場主要原料供應來源之一，在紡織產業中，瑕疵品的出現是不可避免且持續存在的一個問題，尤其是在生產線的過程中有許多機器參數的設定干預到了最終產品的良率，降低了產品的價值和製造商的利潤。 如何有效同時降低多種紡織瑕疵是本研究之重點，目的在於找出對多種瑕疵種類與資料集特徵，本研究對生成型資料集進行數據分析以及規則探勘，選擇並找出對相對應的瑕疵種類有重大影響之特徵集，辨識出每個關鍵特徵的最佳範圍，並且將各關鍵特徵之範圍合成為規則。 此規則能提供給製造工程師當作調整參數的建議，並結合製造工程師之經驗，使最終紡織產品的良率得到改善。本研究的實驗結果闡明，在不同種類的資料集中需要使用不同的方法來進行規則探勘，以求得最佳的規則來改善紡織產品的良率，因此，本研究將會使用兩種不同的方法來對同一資料集進行分析與比較。 ;The textiles of Taiwan serve as one of the primary raw material sources for the global functional textile consumption market. Within this manufacturing, the occurrence of defects is an unavoidable and persistent issue. Particularly, myriad machine parameters during the production line process exert an influence on the final product′s yield rate, subsequently depreciating the product′s value and eroding the profit margins for manufacturers. A central tenet of this study is the effective mitigation of various textile defects. Our objective is to discern patterns among multiple defect types and key parameters or features in the dataset. In this pursuit, we undertook data analysis and do the rule mining on synthetic datasets. After key features that significantly impact corresponding defect types were identified. Each key feature′s optimal range was delineated, and a merged set of rules encapsulating the ranges of these key features was constructed. These rules proffer suggestions for manufacturing engineers to refine parameter adjustments, and when integrated with the engineers′ experiential knowledge, can enhance the yield rate of the final textile products. Experimental results of this study show that different dataset types necessitate distinct rule-mining methodologies to optimize textile product yield rates. Consequently, this research employed two disparate methods to analyze and compare a singular dataset.
