Enhancing mechanical property predictions of woven fabrics: A dual regression approach

Abstract

In today's marketplace, fabric manufacturers strive to attain client satisfactions, which are accomplished by ongoing testing of qualities that impact comfort and quality. The ability to predict these features reduces testing time and expense while maintaining the requisite degree of quality. Consequently, this work aims to develop robust prediction models for key mechanical properties of woven fabric in both warp and weft directions by conducting a comprehensive comparative analysis of predictive methods utilizing statistical modeling techniques, specifically multiple linear regression and multiple non-linear regression analyses. Experiments conducted with various samples of cotton-polyester blends and weft densities form plain woven fabrics, examining properties as tensile strength, bending stiffness, and elongation% in warp and weft directions. Using multiple linear regression to develop six prediction models. Each model was trained on 36 samples and tested on 15 to assess its predictive performance. Comparing model precision allowed for the selection of the most precise prediction model. Based on the MAPE values ranging from 0.010 % to 0.047 %, regression models achieved remarkable precision in the prediction of stiffness, strength, and elongation% in sequence.Thus, the use of these models in weaving mills is therefore highly advised in order to properly forecast all mechanical parameters while reducing testing expenses and material waste.