A Comparison of the Three Techniques in Predicting Breaking Strength of Cotton and Blended Woven Fabrics

With the development of technology, artificial intelligence applications in the textile industry are increasing. The uses of these methods present very good results in cases where statistical methods are lacking in the accurate evaluation and analysis of the past data of the enterprises and the estimation of their future situations. In this study, some models are developed, based on this relationship, to estimate the breaking strength of cotton woven fabrics and polyester/viscose blended woven fabrics separately. Breaking strength is considered one of the most important performance characteristics of woven fabrics. It is mostly determined by the structural elements of the fabric. Multiple linear regression, artificial neural networks and random forest algorithms are employed to perform statistical and stochastic analyses on these elements by using industrial data. A total of 147 fabric data sets in warp and weft directions were used for training and test data in cotton fabrics, and 53 fabric data sets in warp and weft directions in blended fabrics. Appropriate models are generated by using Minitab Statistics and Matlab software. Yarn linear densities, yarn production methods, twist amounts, fabric densities, crimp ratios, unit area weights, various weave factors and fabric structure factors were selected as variables of the models estimating the breaking strength of fabrics in both warp and weft directions. These factors were included in the models separately, and the subset that gave the best results was selected and the models were revised. For the three models created, it was seen that the regression models and models based on artificial neural networks performed well in both cotton fabrics and blended fabrics, while random forest algorithms were not very accurate in estimating the breaking strength.