Utilization and comparison of artificial intelligence techniques for the design of standard woven fabrics

The utilization of artificial intelligence techniques in the textile and apparel sector has been increasing as the adaptation of recent developments in science and technology into industry continues. The use of such methods are particularly very useful when making predictions based on the past company data in the cases where statistical methods are likely to be insufficient. An accurate projection of both structural and performance properties of woven fabrics is extremely important in fabric design. In this study, several models based on multiple linear regression, artificial neural networks and random forest algorithms were developed to predict basic structural properties and performance qualities of woven fabrics. Industrial data from the two groups of woven fabrics, pure cotton and polyester/viscose blend, are used to predict raw and finished fabric width, fabric unit area weight, warp and weft crimps, togehter with breaking strength, tear strength and seam slippage strength, which are the most important performance characteristics. Minitab Statistics Program, Weka and R software were used for creating the relevant models. For the sake of revealing the best relationship between the properties and variables of fabric, various explanatory variables influencing the fabric properties were taken into consideration and several models were developed and the overall results are compared. Among all the models created by four different techniques, random forest algorithms appear to have better prediction ability for the structural properties of fabrics while regression models prove better values for the performance characteristics of fabrics. Keywords: Woven Fabric, Regression, Artificial Neural Networks, Random Forest Algorithm, Mars Technique, Fabric Design.