Recognition of fault process parameters for diameter uniformity variation in melt spinning

Abstract

This study focused on the abnormal condition of diameter variation in melt spinning, and attempted to recognize the machine processing parameter that deviates from the set value. The experiment used polypropylene as the experimental material. The machine processing parameters included nine factors, including temperatures in three extruder barrel sections, temperature of metering pump, die temperature, spinneret temperature, speed of an extruder screw, speed of metering pump and speed of take-up roll. The biaxial laser sensor measured the yarn diameter instantly, and the optimum parameter combination for minimum diameter variation was obtained by Taguchi method and analysis of variance (ANOVA). The degree of influence of various processing parameters on the diameter variation was determined. The optimum parameter combination was used as parameter setting value, and the processing parameters of various factors were changed for experiment, so as to obtain the signals of abnormal condition. Two methods were used for feature extraction. The minimum entropy of wavelet transform signal was used as eigenvector, and two features were selected by analyzing the statistical process control chart, which are skewness and kurtosis. The abnormal processing parameters and normal condition were recognized effectively and accurately by using the three eigenvalues and back-propagation neural network. With enough training samples, the recognition success rate was as high as 100 %. For complete abnormity diagnosis of melt spinning machine, a two-factor classification process was established by using the single-factor classification result and back-propagation neural network. The experimental results proved that the proposed method can recognize various abnormal conditions successfully and effectively.