Determining the quality of shear-viscosity models fitted to experimental data
Sprache des Vortragstitels:
Englisch
Original Tagungtitel:
PPS Ferrol 2024
Sprache des Tagungstitel:
Englisch
Original Kurzfassung:
Polymer processing aims for manufacturing plastics products of sufficient (or highest possible) quality for minimum costs. A prerequisite is process design, which must consider properties and behavior of both the processing machine and the processed material. Therefore, the rheological behavior of polymer melts must be measured and modelled to optimize, for instance, machine operating point or material composition.
The most prominent rheological property is shear viscosity, given as a function of shear rate applied to the material. A standard and laborious technique for obtaining a full shear viscosity model is through point-by-point measurements of viscosity at pre-defined shear rates with plate-plate and HPCR devices, i.e. acquiring parameters that correlate with viscosity like pressure drop and flow-rate, and fitting an analytical model to the data using regression methods.
Our approach is to ease the measurements by digitalization of our on-line rheometer ? a rectangular sensorized slit die at the outlet of a co-rotating twin-screw extruder ? to provide high-and-reproducible-quality models while accelerating and automating data acquisition.
In this study, methods for determining for model fit quality and convergence to experimental data are investigated. A large dataset (>80) of shear viscosities at varying shear rates was measured using the on-line rheometer. A constant-temperature Carreau model was fitted when either randomly increasing the number of points from the measured dataset or applying Bayesian optimization. Comparing to the conventionally determined viscosity curve, each fit quality was assessed by mean error scaled by point count, p-value test, and model parameter convergence.
Through this procedure, we aim to find the minimum number and the optimal distribution of shear rates. If Bayesian optimization outperforms, it will be further used to automate and optimize the data acquisition procedure for better acquisition time and model quality.
Forschungs-