The Neural Network Prediction Model Of Rotor Spinning Yarn Properties

As an open-end rotor spinning method, rotor spinning is a new spinning technology, which has gotten widespread industrial application among different new kinds of spinning technology. In the rotor spinning process, the fiber properties, features, rotor spinning process parameters and so forth have a major impact on yarn properties, and they have non-linear with yarn quality. Neural network can predict the propeties of yarn, There will be too many input variables if all factors are considered when use the artificial neural networks to predict the yarn properties of rotor spinning. It has been found that not all the factors can effect yarn properties. In fact, the number of the samples in the spinning mill is limited, Therefor, it's necessary to establish ANNs model to accurately predict the quality of rotor spun yarns with less input variables. Empirical approach and certain mathematical method have been using in selecting input variables. The former is subjective then cannot show the importance of how the input variables influence the output variables; the later may lose its justness because different mathematical methods focus on different aspects.This paper introduces six methods to rank input variables: the first one is Principal component analysis, analysis into the change which output influced the input useing the calculation method of similarity coefficient; the second one uses a data sensitivity criterion based on a distance method that analyzes the measured data of raw material, machine , process and rotor spun yarns; the third one takes the human knowledge on the rotor spun yarn into account; the forth one is Fuzzy cluster; the fifth one is is a new fuzzy selection criterion which thinks much of the sensitive of the measured data; the last method is grey incidence analys.First of all, the research chooses 17 input variables from raw material properties, spinning machine and spinning process. Then there are experiments on raw material properties, spinning and rotor spun yarns properties. Due to the experiments, all input/output data is gained. After that, the six methods are use to rank the input variables for certain property of the yarns and six different rankings come out. At last, the fuzzy selecting criterion plays an important part again in combination the rankings.The ANN model is established for every specific property of the yarns, at first, the model is taken to test and verify the efficiency of the selecting methods; then, from the first to the seventh variables in the ranking line are taken as the most relevant input variables of the ANN for predicting the quality of the rotor spun yarns.The result is that, the prediction of the ANNs is close to the real data. It's proved the selecting methods are efficient and right, as well as the ANNs can be used to predict the quality of rotor spun yarn.