Parameter optimization in melt spinning by neural networks and genetic algorithms

Abstract

An approach for determining parameter values in melt spinning processes to yield optimal qualities of denier and tenacity in as-spun fibers is presented. The approach requires a fewer number of experiments than conventional methods. An orthogonal array in the Taguchi method determines the minimum number of experiment trials to be conducted. Whether the experimental data are adopted to train a neural network is justified by an analysis of variance(ANOVA) and confirmed by experiments. A neural network relating 11 process parameters and two quality characteristics is constructed. The genetic algorithm is aimed at finding parameter values in a continuous solution space to optimize a performance measure on denier and tenacity qualities, based on the neural network. The performance measure is evaluated by the technique for order preference by similarity to ideal solution (TOPSIS). To expand the solution space, three different sets of level values for the orthogonal array are chosen from the ranges where the melt spinning will properly work. The results demonstrate that the proposed approach gives the smaller denier and the larger tenacity of polypropylene(PP) as-spun fibers than the Taguchi method.