Arkun - Georgia Tech - 9522564 Nowadays chemical plants use the same equipment to produce a wider variety of products. For higher productivity and better utilization of resources, the equipment must be easily adaptable to different operating regimes resulting from changes in raw materials, product specifications, market demands, different energy sources and other external disturbances. Production of undesired materials, loss of product quality and unsafe operation are attributed to control problems encountered during long periods of plant transitions between different regimes. The objective of this research program is to investigate the fundamental issues related to the design and implementation of transition control systems. The PI plans to develop and test new theoretical approaches, and transfer the theory into useful industrial practice. His approach stems from a unified treatment of certain modeling, monitoring, control and plant-wide optimization issues underlying the transition control problem. It is often difficult to develop a single first principles model which describes the plant dynamics over all the operating conditions; neither is it practical to identify such a model from input-output data. Therefore, quantitative tools are first proposed to partition the plant operating space into local model descriptions that are validated at the individual regimes. Next these models are combined into a global model in order to explain the plant behavior in transition regions as well. Different global modeling techniques are planned, including model matching, output weighting, linear parameter varying systems (LPV) and neural nets with self-organized learning. Based on recent model validation results, a method is given to quantify the uncertainty associated with these global models. The transition controllers will be designed based on the global model and have the following properties: (1) they are robust control laws which guarantee a certain level of performance during transiti ons, and (2) they reduce to local regulatory controllers, when the plant operates in its isolated regimes. For design purposes different approaches will be studied including robust scheduling and predictive control using LPV or self-organized neural net models. The implementation of transition controllers within the context of plant-wide control will be tackled. This will include interfacing with the lower level feedback loops (PIDs) and coordination with the higher level optimization task where scheduling of transition objectives is performed. During the course of this research the PI will collaborate with the research group at DuPont (this is a GOALI project). It is anticipated that the control methodology will help alleviate the generic transition problems encountered in industrial processes and improve plant operations.
