The steady growth of the plastics industry during the last decades has driven development in
polymer processing with the extruder as the most important machine for producing pellets in the
production of polymer materials as well as in the production of sheets, films, pipes, profiles,
coatings or cables. To meet the ever-increasing demands on polymer products and plastics
machinery, polymer processing requires further analysis and optimization using most modern
techniques. Therefore, this thesis combines both numerically driven modeling and experimentally
driven modeling by different data mining techniques in order to build up regression and
classification models. These modeling techniques enable knowledge discovery from data as well
as the development of several regression and classification models that can serve as a basis for
predictive modeling and digital twins.
Numerically driven modeling is applied to describe the throughput-pressure gradient relationship
for single-screw extruders as well as for woven screens used for melt filtration. The mathematical
equations developed, enable fast and stable computational modeling without the need for further
A novel, heuristic method for modeling the 2D dimensionless throughput-pressure gradient
relationship of non-Newtonian polymer melt flows in single-screw extrusion is presented.
Extensive parameter studies are performed varying the influencing variables and numerically
solving the set of coupled, non-linear, partial differential equations by the shooting method. These
data serve as a basis for regression modeling. Therefore, different generally valid analytic
equations are developed describing the 2D throughput-pressure gradient relationship considering
the shear-thinning behavior of the polymer melt. These novel approximation methods are
numerically confirmed by a performed error analysis.
Heuristic optimization algorithms are further used to develop generally valid analytic equations for
estimating the initial pressure drop of woven screens used in polymer processing and recycling.
Based on defined elementary cells for square woven and Dutch weave screens, extensive
numerical modeling is performed creating a huge data set of independent design points by means
of parametrization. Symbolic regression models are derived by first analyzing and simplifying the
governing equations and then transforming them into dimensionless form applying the theory of
similarity. The high accuracy of the developed mathematical models is numerically confirmed and
Experimentally driven modeling in combination with data mining is applied to identify the most
relevant process parameters influencing the quality of final applicate powder coatings.
Experimental studies are performed on co-rotating twin-screw extruders processing different
powder coatings with varying screw configurations and process conditions. Statistical methods
and regression, as well as classification techniques, are applied to develop specific up-scale
models that can further be used for in-line process and quality control.