Modeling the Properties of Core-Compact Spun Yarn Using Artificial Neural Network

In this research, the compact-core spun yarns have been produced using RoCoS roller and the effects of filament pre-tension, yarn count and type of sheath fibers were investigated on the physical and mechanical properties of produced yarns such as strength, elongation percentage, hairiness, and abrasion resistance. After statistically analysis on the obtained results, for modeling the core-compact yarn properties, the regression and artificial neural network (ANN) were used to predict the physical and mechanical properties. Trial and error method was considered for determining the best of ANN topology. For this aim, 1110 topologies of ANN (with different hidden layers and neurons in each hidden layer) were investigated for each property. Moreover, to evaluate the accuracy of the created ANN three indexes were used, namely mean absolute percentage error (MAPE), mean square error (MSE), and correlation coefficient (R-value). It was observed that the most accurate results were obtained based on MAPE and the best topology for predicting all properties is a two-hidden layer ANN (maximum MAPE < 0.10) except for the abrasion resistance which is a three-hidden layer ANN (MAPE < 0.17).