CC-03-162 An Essential Evolution in Nonliner Polymer Production Control: An Industrial Case Study

Description:

There are many objectives in the optimization of commercial polymer manufacturing processes. The economic incentives are often to reduce variability, increase throughput and reduce off-specification material produced during product transitions. An additional “hard constraint” is obviously the integrity of process operations and plant operational safety. Process transitions, apart from start up and shut down, are the most dangerous procedures performed on a polymer plant. Controller integrity during these changes is therefore essential. It was first thought that neural network-based solutions could provide a way forward; however, it is now clear from the literature (Zakrzewski, [8]) that guaranteeing the safety of an accurate neural network online would be infeasibly expensive in a manufacturing environment. This paper will present the commercial application of an alternative universal approximating technology (The State Space Bounded Derivative Network) with many globally-guaranteed properties that make it an intrinsically safe solution within the framework of a nonlinear model based predictive control strategy. The result of this breakthrough is the application of the world’s first truly universal, nonlinear model predictive controller which has delivered record breaking transition times on a commercial polyethylene production facility.