Modelling ring spun yarn properties using general regression neural network.

Abstract

Most textile firms use the ring spinning system to spin yarn because it produces strong yarn. However, it is expensive due to extra work and labor e.g. carding, combing, drawing and roving formation. Also, the machine setting parameters can affect the quality of yarn and therefore there is need to test the yarn quality. Textile firms have therefore used. several methods and machines to test for yarn quality e.g. visual examination, cut and weigh methods, gravimetric method, uster Technologies etc. Different instruments have been used e.g. Oster technologies, wrap reel to measure yarn length, Analytical Balance, Knowles Balance and Quadrant Balance to determine yarn count etc., However, these cannot predict yarn properties. Textile spinning firms are thus faced to deterioration in their research capabilities in the last years due to failure of the present technology in predicting yarn properties, otherwise, they can use try and error method which increases the cost of production if results obtained are poor. The aim of this work is to model and predict the ring spun yarn properties (strength, evenness and imperfections). Yarn was therefore obtained through a series of experiments carried out at Fine spinners (U) Ltd (FSL) in Kampala-Uganda. Yam produced was used in developing a General Regression Neural Network (GRNN) to probe the yarn properties of 100% cotton. This was done by the ring spinning system and the parameters i.e. yarn count, yarn twist and spindle speeds were used as inputs for the GRNN model. The same parameters were used as inputs for the linear regression and the results compared to validate, the GRNN model. According to the results GRINN had better R.2, RMSE and SSE therefore rendering the GRNN model a success and superior to linear regression models in predicting yarn properties of' 100% cotton.