Diameter prediction and optimization of hot extrusion-synthesized polypropylene filament using statistical and soft computing techniques

Abstract

In this study, statistical and soft computing techniques were developed to investigate effect of process parameters on diameter of extruded filament made of polypropylene in hot extrusion. A multi-factors experiment was designed with process parameters of screw speed, roller speed and die temperature. According to the design matrix, twenty four experiments were conducted. The diameter of the extruded plastic filament was measured in each experiment. Subsequently, statistical analysis was used to identify significant factors on diameter of extruded filament. Predictive models of response surface methodology (RSM) and radial basis function neural network (RBFNN) were applied to predict the diameter of extruded filament. The optimal process parameters to maintain the diameter of the filament closest to the target value were identified using the cuckoo search algorithm (CSA), and particle swarm optimization (PSO). Performance analysis demonstrated the superior predictive ability of both models, in which the prediction errors of 0.0245 and 0.0029 (in terms of mean squared error) were obtained by RSM and RBFNN, respectively. Considering the optimization methods, the optimization approaches of using CSA and PSO were promising, in which average relative error of 1.28% was obtained in confirmation tests.