Tekstil sektöründe veri mühendisliği ve yönetimi

Recently, enormous amounts of data are generated every day in textile industry. These multivariable and nonlinear data include raw material characteristics, machine settings, process parameters and quality attributes of the textile product. Deriving useful patterns and valuable knowledge from these raw data provides making right decisions to increase quality and productivity for textiles. To serve the purpose, this thesis focuses on the application of the data mining and machine learning techniques in the textile sector. Data engineering is a discipline that concerns with data mining techniques for data processing and analysis. Data mining techniques can be grouped in three main categories: classification, clustering, and association rule mining. In this thesis, several case studies were conducted for each category. In the classification-based case studies, ensemble learning methods were proposed to improve prediction performance in textile sector (i.e. to determine stab resistance performances of knitted structures) as well as deep learning methods for textile object identification. As a clustering-based study, a novel hierarchical clustering approach, named k-Linkage, was proposed that calculates resemblance between pair of clusters considering k samples from two clusters. In the association rule mining study, an extended FP-Growth algorithm was used to discover the relationships between yarn and fabric properties. In the thesis, several experimental studies were performed for each study to demonstrate the performances of the proposed methods. In each experiment, the proposed approaches were applied on real-world textile data and compared with the existing approaches in terms of different evaluation measures. In general, the results obtained from each experiment indicate that the proposed approaches in this thesis achieve more accurate results than the conventional solutions.