Application of Principal Component Analysis and Neural Networks in the Determination of Filler Dispersion during Polymer Extrusion Processes

Mineral filler-reinforced polymer is an important family of polymers designed to achieve high mechanical impact strength. The state of mineral filler dispersion in a polymer matrix strongly affects the mechanical properties of the product and is an important piece of information for the extrusion-based fabrication process. In this work, a measurement system composed of 2 ultrasonic sensors, 3 pressure sensors, a thermocouple, and an electric current meter of the extruder motor drive were used to monitor the extrusion of a calcium carbonate powder-filled polypropylene system. Three principal components most correlated to the state of filler dispersion were extracted from the data set collected by the multiple sensors and fed as inputs to a neural network model designed to determine the dispersion state of the filler. By using this approach, we were able to achieve an accuracy of better than 0.05 on the estimation of dispersion index. This work has demonstrated the feasibility of combining our multi-sensor monitoring system with principal component analysis and neural networks for on-line determination of mineral-filler dispersion in polymers.