Development of an inference based control scheme for reactive extrusion processes

Reactive extrusion processes are widely employed in polymer processing industries for the production and modification of commercial polymers (e.g. ethylene co-polymers, polyamides, etc.). Due to strongly interacting process mechanisms and the non-availability of measurements of quality variables, q (melt index, viscosity, density, etc.), as well as product end-use properties, w (toughness, UV/chemical resistance, etc.) at a high enough frequency, it has been very difficult to guarantee acceptable end-use product performance for this class of processes. To accomplish the challenging goal of assuring acceptable end-use performance, we have developed a modeling and control framework for effectively controlling key product properties and meeting customer requirements on end-use performance. In this framework, an appropriate quantitative representation of the relationships between variables across the entire processing chain is realized through a modeling scheme consisting of a hierarchical network of judiciously selected models. To address the issue of the non-availability of frequent product property measurements, two models---one each for q and w---are used for inferring the properties at a fast enough rate to control these properties online. The models constitute an integral part of a control scheme consisting of a fast model-based controller ( C1) for the inner loop between the manipulated variables (screw speed, feed-rates, etc.) and the process outputs (die pressure, melt temperature, etc.); a slower model-based controller (C2), which translates end-use performance objectives to set points for the process outputs; and a customer feedback regulator (C3), which translates customer feedback on product performance to appropriate product property targets so that the product meets its end-use requirements precisely.; The work presented in this thesis was aimed at developing the models and controllers of the scheme for an example reactive extrusion process and experimentally implementing the overall scheme on the pilot-scale process. An experimental study of process dynamic behavior was conducted to gain deeper insight into the role of the process mechanisms in developing product quality and end-use properties. We found that while reaction and flow were the most critical mechanisms affecting the product quality, melting played the most dominant role in the evolution of the product end-use properties. Such insight was used in developing a fundamental quality inference model and a fundamental/empirical (hybrid) end-use property inference model.; The task of designing the model-based controller C1 was challenging because of strong process nonlinearities, and process ill-conditioning at an operating point in the low viscosity regime. Specially designed input signals were used to identify a reasonably accurate empirical model of this ill-conditioned process, and C1 was designed as a gain-scheduled, model predictive controller that can function effectively in the presence of process nonlinearities. The outer-loop controller C2 was also implemented as a model predictive controller, and the performance of the overall scheme was evaluated in simulations. We found that the performance of the scheme was satisfactory in achieving product property specifications that are physically realizable with the current system configuration.; Customer feedback on product performance is the true indicator of whether the product was able to meet its end-use performance requirements; this feedback has been completely missing from the control schemes that have been traditionally designed not just for reactive extrusion processes, but for all other polymer processes. We developed a novel procedure for implementing customer feedback control. First, we developed a modeling method for relating customer feedback---which is usually in the form of a binary variable that quantifies the product performance as "acceptable" or "unacceptable"---to real-valued product...