Predictive Control of Rubber Mixing Process Based on Neural Network Models

The work presents an attempt to establish a closed-loop control system for mixing rubber compounds in internal mixers by means of soft computing methods in form of neural networks and methods of information theory. As the basic physical aspect to be controlled is chosen the course of compound viscosity during the mixing process, since it comprises both the features of the mixed materials and those of processing. Besides, viscosity is proportional to the rotor torque which is a continuously measurable quantity. Since too many factors influence viscosity, or rotor torque, to be modeled analytically, neural networks, together with information theory, have proved to be suitable method to create efficient means for viscosity course redirection in the case of deviations, some of which being unavoidable despite measures. Thus an on-line predictive control system is formed, enabling compound production of more uniform quality.