Control-relevant model identification of reactive extrusion processes

In reactive extrusion processes, where process dynamics are determined primarily by the strong interactions between fluid mechanics, reaction kinetics, heat transfer and the complex extruder geometry, effective control of product quality and end-use properties by classical techniques is virtually impossible. Under such circumstances, effective control is facilitated by the application of model-based control, which requires the availability of process models that are suitable for control system design and implementation. As a result of the complexity of the fundamental mechanisms underlying the typical reactive extrusion process, first-principles process models are impractical for control, making system identification from judiciously acquired input/output data a more practical alternative. In this paper, we present and illustrate a strategy for identifying appropriate control-relevant models for this class of processes. The strategy consists of carefully designed identification experiments involving preliminary process tests and a final test, followed by judicious model structure selection, parameter estimation and model validation. Practical application of the procedure is demonstrated on an experimental system involving the reaction of a functionalized ethylene co-polymer, “Elvaloy” (Ethylene/n-Butyl Acrylate/Glycidal Methacrylate Terpolymer) with an acid co-polymer “Nucrel” (Ethylene/Methacrylic Acid Copolymer), in a Coperion W&P ZSK-30 mm co-rotating, intermeshing twin screw extruder.

► We presented (and demonstrated experimentally) a technique for obtaining control-relevant models for reactive extrusion processes.
► Empirical models are often the best options for model-based control of reactive extrusion processes.
► Identification of control-relevant models for these processes requires a two-stage procedure of preliminary process tests to gather information for the final test.
► The global nonlinear characteristics of reactive extrusion processes can be efficiently represented by a judicious combination of multiple linear models that are valid in different operating regimes.