PurposeThis study aims to create a classification system enabling users of 3D virtualization software to intuitively perceive the drapability of fabrics.Design/methodology/approach1,001 fabrics were used, and thickness, bending property, and tensile strength were identified as main mechanical properties influencing drapability; they have been set as independent variables in the model established to predict drape coefficient.FindingsA system to classify fabrics into eight groups by drapability was suggested by a cluster analysis, and a multinomial logistic regression analysis was used to set a model that allows users to predict which group a fabric belongs to from its mechanical properties.Originality/valueThis paper provided basic materials for the construction of a virtual clothing simulation system, which is believed to contribute to cost and time savings in decision-making by reducing the number of trials and errors required by the conventional approach.

Cutting-edge technology is being used in the fashion industry for three-dimensional (3D) virtual fitting programs to meet the demand for clothing manufacturing as well as textile simulating. For expanding the textile choices of the program users, this research looks at the indexation of tactile sensations, the texture of fabrics, which has been subjectively evaluated by the human hand. Firstly, this study objectively measured and indexed the surface smoothness and fiber softness of 749 fabrics through a tissue softness analyzer that mimics human hands. Secondly, after statistical analyses of the drape coefficient, each bending distance and Young's modulus for the initial tensile strength in the warp–weft directions, the thickness, and the weight of the fabrics, it was found that drape (Pearson coefficient = 0.532) and bending properties are the key factors in the fabric surface smoothness (TS750), while the fiber softness (TS7) showed a weak correlation with thickness (Pearson coefficient = 0.364), followed by the log value of the Young's modulus in the weft direction. Thirdly, we classified nine clusters for TS750 based on the 11 regression variables with significant Pearson coefficients, and characterized each cluster in order of surface smoothness (TS750) after Duncan post-hoc tests and analyses of variance (all statistically significant, p < 0.01) with microscopic surface images of one sample for each cluster. For precise TS750 classification, we finally trained the 267 samples with the same 11 variables, resulting in 93.3% prediction through an artificial neural network with multiple hidden layers. This prediction with Fisher discriminants for the clusters will enable the 3D virtual program users to predict further clustering of newly added fabrics.

No abstract