Design Of Positive Draw-back Motion Of Wool Spinning Frame And Comparison Of Prediction Models Of Worsted Yarns Performances

Rabbit hairs have advantages of light-weight, smoothness and good-brilliancy. And the products have virtues of beautiful appearance, bright color, soft touch, fluff standing and wear light, which are popular among people. However, flake open-angle of rabbit hair fibers are small, each flake tightly compress, and boundaries between flake are insufficient distinct. Meanwhile, short length, little crinkling, shallow peak, even without curvature are also characteristics to rabbit hairs. The above features lead to small cohesion of rabbit hair fibers, hard to be yarns. Much research has been done to optimize the process lines of spinning high counts pure rabbit hair worsted yarns by various enterprises. And the mean fiber length, mean fiber diameter and other indications have achieved the standard of spinning high counts pure rabbit hair worsted yarns to same extent. However, the enterprises still need to face a problem, owing to unwinding method of rove is passive in wool spinning frame; twit and broken ends of rove could easily take place in part of suspended flyer to traverse rod when spinning high counts pure rabbit hair worsted yarns. And it is harmful to spin high quality high counts pure rabbit hair worsted yarns.Traditionally, the method of single-tout test is used to determine the process parameters in many enterprises, which causes waste of plenty of manpower, material and time. Then, a number of prediction methods based on mathematical models and empirical formula are established. However, the prediction results are not accurate. Neural networks have a powerful self-learning ability and structural variability in the process of solving problems, which are needn't accurate mathematical models and quite suitable for mills to predict the performances of yarns. And little deeply work has been done to study the prediction models of artificial neural networks of yarns performances under the condition of large-scale input samples and high input nodes.This paper designed a positive draw-back motion used in wool spinning frame, which could avoid twit and broken ends of rove that existed between the suspension flyers and traverse rod. The principle of the motion is to achieve micro-tension or zero-tension of the rove in the rove bobbin rotated, and make the linear surface velocity of rove lying in rove bobbin surface equal to that of later roller of draft field, which has contribution to reduce breakage rate and unevenness rate of rave that improving the quality of high counts pure rabbit hair yarns. And breaking strength, unevenness and elasticity of the worsted yarns with and without positive draw-back motion were tested and made a comparison. Experimental results show those individuals'performances of the high-count pure rabbit hair worsted yarns of positive draw-back motion could improve to same extent, which validated the effect of the motion.BP and RBF of artificial neural networks were used to predict the performances of worsted yarns(not only including high counts pure rabbit hair worsted yarns produced by wool spinning frame with positive draw-back motion, but also containing wool worsted yarns, cashmere worsted yarns and other types of yarns). Prediction performances of different artificial neural networks were analyzed, and optimal models for plants were determined. First of all, one and two-hidden layer BP neural networks were used to predict unevenness and breaking strength of worsted yarns on under the condition of large-scale input samples and high input nodes, prediction performances of BP neural networks were analyzed from the aspect of prediction accuracy, stability and training steps. Experimental results show that two-hidden layer BP neural network model with 9 hidden layer nodes is the preferred model for predicting the unevenness of worsted yarns, and the relative coefficient value reflected prediction performance is 0.9849.Two-hidden layer BP neural network model with 9 hidden layer nodes is considered as the preferred model while predicting breaking strength of worsted yarns,and the relative coefficient value is 0.9888. Both of the relative coefficient values of two preferred models chosen are very close to 1, which means BP neural network has higher prediction performances under the case of large-scale input samples and high input dimensions.In addition, RBF neural network also had been used to predict unevenness and breaking strength of worsted yarns on under the case of large-scale input samples and high input dimensions, and prediction performances of BP and RBF neural networks were comprehensive evaluated based on training steps, difficulty of modeling and prediction abilities. Experimental results show that training speed of RBF neural network is faster than that of BP neural network on the condition of high input dimensions. But the spread value of RBF neural network could only been determined by using a certain amount of experiments, and abnormal samples should be excluded when using RBF neural network to predict performances of worsted yarns. Therefore, it is more complex to use RBF neural network to establish model. The relative coefficient values of unevenness and breaking strength are 0.9438 and 0.9423, respectively, after excluding abnormal samples step. The prediction abilities of BP neural network are slightly higher than that of RBF neural network. In summary, comprehensive performances of BP neural network are slightly superior to that of RBF neural network under the case of large-scale input samples and high input dimensions. And BP neural network is more suitable for plant technicians to predict quality of yarns.