The purpose of this project is to determine the feasibility of applying the Parallel Distributed Processing (PDP) approach to on-line feedback control of nonlinear chemical processes. PDP theory assumes that many complex processes in cognition can be represented by a large set of individually simple interconnected units with complex interactions. Parallel refers to the fact that units excite and inhibit each other simultaneously in the network while distributed indicates that knowledge consists of connection strengths between pairs of units that are distributed throughout the network. This approach "learns" as it goes along, i.e. it is a technique that mimics human intelligence gathering. Adaptive controllers work by changing in response to process changes. This is feasible for linear systems but much more complex for nonlinear systems. In this research project, the PDP network model will be used for the development of sensitivities and local linearizations that can be used for controller design for nonlinear chemical processes. The PI plans experimental verification of the algorithms developed on a laboratory methanol distillation column. This will permit comparisons with self-turning and other adaptive controllers.
