The Neural Network Prediction Model For The Yarn Quality Of Worsted Spinning

In the process of worsted spinning,the fiber properties,spinning elements and spinning process have great effects on the yarn quality,and there are nonlinear relationships among them.The prediction of yarn quality and inversion of spinning parameters can be realized with a neural network.But using miscellaneous factors as input variables will decrease the network operation efficiency and prediction accuracy.In fact,not all factors have significant effects on a yarn quality index.In practical production,the number of collected samples is often limited.Therefore,it is necessary to select the input variables to establish the neural network prediction model for the yarn quality.It is also necessary for the design of spinning technology to establish the neural network inversion model.In this thesis,according to the practical production,seven spinning parameters which affect yarn quality from the aspects of the fiber properties,spinning elements and spinning process are determined.They are the fiber diameter,fiber length,traveler number,draft ratio,spindle speed,yarn twist and yarn fineness.The input and output variables of the neural network are obtained by the spinning test and yarn quality test.Five input variable selection methods are used:(1)an input-output effect based selection method;(2)a expert knowledge based selection method;(3)grey correlation analysis method;(4)TOPSIS comprehensive evaluation method;(5)Fuzzy inference method.By using the fuzzy inference method,the results of the five selection methods are combined and final ranking results are obtained.Based on the ranking results of the input variables,the BP neural network prediction models are established by taking the spinning parameters of ranks 1-4 and ranks 4-7 as the input variables.It is found that prediction errors of the input variables of ranks 1-4 is smaller than those of ranks 4-7,indicating validations of the input variable selection results and the combined results.Then the spinning parameter of ranks 1-4 are selected as the input variables to establish neural network prediction model for the yarn quality.Seven yarn quality indices are used as the input variables to establish the neural network inverse model for the seven spinning parameters.The results show that the prediction and inversion results are close to the measured results,which proves the neural network models are effective.In view of the characteristics of the genetic algorithm and BP algorithm,the genetic algorithm is used to optimize the structure of BP neural network.Then GA-BP neural network prediction and inversion model for the yarn quality of worsted spinning are established.The results shows that the mean values and standard deviations of the prediction and inversion errors of the optimized neural network are smaller than those of the original neural network.This research confirms that the neural network based on genetic algorithm is feasible and effective for predicting the yarn quality and inversing the spinning parameters.In this thesis,the neural network prediction and inversion models for the yarn quality of worsted spinning based on the input variable selection and genetic algorithm are established.The results can provide a theoretical reference for improving the worsted spinning process and equipment and the yarn quality of worsted spinning.