There is a rising need for high-performance, multi-variable control in the chemical process industries to enable corporations to maintain a competitive edge during rapidly changing market conditions and increasing concerns for energy conservation and environmental protection. These new controllers will have to deal with the following factors: (1) more control problems involving highly interactive processes with a large number of inputs and outputs will result from increased plant integration; (2) more control problems will involve constraints since the economic optimization forces the plant to be operated close to the intersection of constraints; and (3) control problems will embody more nonlinear/time- varying aspects because of frequently changing operating conditions and increased number of sensors and actuators. Three serious issues arise for designing control systems for large-scale, interactive processes subject to constraints: ensuring robustness to model uncertainty, minimizing the effects of constraints and monitoring and maintaining performance of the control system in the face of a changing plant environment with minimal human intervention. The objective of this project is to develop an integrated recursive-identification/control technique that is needed to generate high-performance, multi-variable controllers with the capability of adapting to changes in the environment. A recursive identification algorithm that provides a family of possible models and a predictive control algorithm minimizing the "worst-case" error for all possible models are to be developed. The recursive identification algorithm is to be equipped with the intelligence to reset the model parameters and uncertainty bounds appropriately from on-line data when a sudden change in the plant dynamics occurs. Together they comprise a multi-variable, self-tuning control algorithm that can maintain high and robust performance despite little a priori information on the plant dynamics and changing operating conditions. A novel model uncertainty quantification method based on the "finite impulse response (FIR)" model structure is to be employed. The research effort will span a wide array of system and control technology including identification, robust control, predictive control, robust control and artificial intelligence, and involve a comprehensive blend of theory, experimentation and simulation.
