Comparison of Stochastic Identification Techniques for Dynamic Modelling of Plastics Extrusion Processes

Abstract

The important plastics extrusion process is briefly described and the difficulties of modelling the process from physical considerations are outlined. A number of stochastic process identification techniques, i.e. correlation, spectral analysis, generalized least squares, instrumental variable, correlation matching, maximum likelihood and Box-Jenkins algorithms are briefly reviewed. The results of experimental work carried out on a laboratory plastics extruder, using random perturbations in screw speed, are presented. From a comparison of the results of different identification methods, it is suggested that although correlation and spectral analysis techniques can help in improving the understanding of the process mechanisms, the type of models best suited for high level feed-forward computer control are of the Box-Jenkins and maximum likelihood structural forms.