![]() | Only 14 pages are availabe for public view |
Abstract The quality attributes of apparel fabrics and their processing performance during garment manufacture are closely related to basic mechanical and physical properties of these fabrics. The present study is concerned with the prediction of the processability of apparel fabrics from their pre-measured mechanical and physical properties. Two woups of fabrics (cotton shirting fabrics and wool suiting fabrics) were used for the study. Experts from the textile and apparel manufacturing industry have subjectively evaluated the fabrics (of group I) for eight different processability parameters, while the fabrics (of group II) have been evaluated for five different processability parameters according to ”Ito” control chart. The measurements of mechanical and physical properties were carried out by three different techniques: FAST system, conventional lab testing, and Kawabata (KES-F) system. A predictive model has been designed based on artificial neural network; ANN (feed forward ANN with back-propagation) to predict the processability of fabrics under study. The fabric properties represented the ANN input vector, while the processability parameters represented the ANN output targets. The training process of the network reached the fully trained status with all kind of fabric properties measurements. The network accuracy level for processability values of the fabrics of group I was 90.5% with FAST parameters and 93.7% with lab testing parameters. The network approximation for the fabrics of group II was 99% with KES-F parameters. A statistical technique (multiple regression) was used to compare the accuracy of ANN predictive model. A very low accuracy level was found when applying the multiple regression technique (about 32%) in comparison with the ANN results. Practical evaluation of the processability parameters of six new fabrics - based on the actual manufacturing processes - has been carried out and taken as reference to test the accuracy of the ANN predictive model. A percentage error below 13% was found for the prediction of process ability of these fabrics, which indicates a good reliability of the predictive model. |