Estimation of the elongation, tensile strength and crimp values of polypropylene BCF yarns by using artificial neural network technique: An application in textile sector

Nowadays, polypropylene (PP) yarn which is the most consumed fiber as a pile yarn in carpet production is demanded to be softer, brighter, to maintain appearance and to improve its resilience. Although it is known that PP yarn has a disadvantage in terms of softness and brightness compared to natural or other synthetic yarns used in the production of machine made carpets, it has been observed that these properties have been improved with the properties of PP yarn and production parameters. In this study, it is aimed to estimate between PP BCF (Bulked Continous Filament) yarn production parameters and high commercial value properties such as crimp, tensile strength and elongation. In addition, these properties in the determination of the production components, it is aimed to make much more serial and high accuracy production optimization than the traditional methods described as trial and error. In this context, the production parameters of the PP BCF yarn produced by Kartal Halı in 2017-2018 and the elongation, tensile strength and crimp output values of the yarn were estimated by using the Feedforward back-propagation network, Elman network and the Cascade back propagation network from the artificial neural network models. The performance of the ANN models was compared with the coefficient of determination (R2) and the percent accuracy statistical performance indicator values (MSE, MAPE, MAE, RMSE, MSPE, MPE). For the training set, 108 ANN models were created and nine best ANN models (FeedForward14-15-16, CascadeForward13-14-15, Elman1-2-3) were determined. In order to confirm the differences between the measured values of the test data of PP BCF yarns and the measured values of the ANN models, t test analysis was applied and the best three ANN models (FeedForward15, CascadeForward15, Elman3) were selected according to the statistical performance values. The elongation, tensile strength and crimp estimation values of the yarns in these three models were determined. R2 values of the training set of the feed forward back propagation network (FeedForward15) model which gives the best results are 0.8954 – 0.7992 – 0.9047; For the MSE value, 53.65 – 0.33 – 0.25; For MAPE values, values of 10% - 20% - 9%, respectively; 0.9124 – 0.8687 – 0.7919 for the R2 values of the test set, respectively; The MSE value was found to be 57.12 - 0.26 - 0.22, respectively; The values of MAPE were 11%, 46% and 8%, respectively. As a result of these studies, it has been found that the Feedforward back-propagation network model statistical performance values give slightly better results than Elman network and the Cascade back propagation network models.