Neural networks for process control of flat film extrusion systems

Abstract

Focuses on the improvement of an existing nonlinear control system with neural nets. Crucial for the successful exploitation of the improvement potential is the neural networks' outstanding capability for online adaption and the possibility of combining neural networks with existing physical or mathematical models. The primary result is an improved process modelling which allows process model adaption. Using the described capabilities, a predictive feedforward control/feedback control (FFC/FFB) has been developed to overcome the greatest problems in processes with long delay times. The corresponding installation is a flat film extrusion system to produce polymer sheets. Simulation results point out that optimized process control in flat film production can lead to significant material savings as well as significant improvements in product quality.