PREDICTION OF WORSTED YARN BY USING NEURAL NETWORK

摘要

Based on the spinning theory and neural network technique, seven indexes of yarn quality and spinning performance are modeled for prediction purpose. These indexes are yarn evenness, thin places, thick places, yarn tenacity and its variation, elongation at break and the rate of ends-down. Due to the complex interaction on each other between yarn properties, spinning parameters and end-down rate, the compound neural network model is specially employed to predict spinning ends-down. The models are trained and tested by using industrial spinning trial data from a worsted spinning mill, and satisfied prediction results are obtained. Except the rate of ends-down with the square of correlate coefficient (R~2) between predicted and measured values of 0.84, the value of R2 for the other yarn property indexes are all above 0.9. It shows the neural network provides a powerful tool for worsted yarn prediction.