Predicting Yarn Tenacity: A Comparison of Mechanistic, Statistical, and Neural Network Models

Prediction of yarn properties from fibre properties and process parameters is a well-researched topic. For a number of years, mechanistic and statistical models have primarily been used to tackle the problem. Over the last ten years, neural networks have been used in increasing numbers for this purpose. However, a comparative assessment of the performance of these three approaches has not been forthcoming. In this paper, all the three models have been applied on the data available to validate the mechanistic model described by Frydrych (pertaining to cotton yarns). The exercise was repeated for data pertaining to yarns spun from polyester staple fibre in the laboratory. The results conclusively prove the superiority of neural networks over mechanistic models and simple regression equations for predicting ring yarn tenacity from fibre properties and process parameters.