Control structure synthesis and neural network based control for chemical processes

Abstract

In Chapter 2, a new method for dynamic structural transformation for the analysis of distillation control structures is presentd. Since the proposed method is based on the relationships between the block elements of the complete open loop transfer function as structural transformation there is no need to reestablish input relationships at every transformation. Using this framework we show clearly why Haggblom and Waller``s method is valid for the specific cases. For the more general cases including the effects of inventory dynamics, a simple mothod is suggested. Usefulness of the method is illustrated by an example. In Chapter 3, new interaction measures for control structure synthesis are presented. First, all possible interactions in the closed loop system are rigorously analyzed. Based on the analysis, new interaction measures-Type I and Type II Relative Output Sensitivity Matrices (ROSM) and Type I Type II Relative Input Sensitivity Matrices(RISM) -which can measure corresponding actual interactions in the closed loop system are proposed. These new interaction measures can be used to assess directly the true closed loop performance. Relationships between Type I asymptotic ROSM and system stability are established. Also the nonsingular value based interaction measure(NSV IM) which can predict the system stability is proposed. A systematic methodology using the new interaction measures is proposed for analysis and synthesis of decentralized control ststems. Examples are presented to illustrate the significance and usefulness of the proposed method. In Chapter4, a neural controller for process control is proposed that combines a conventional PI controller with a neural network. The proposed neural controller is applied to distillation column control. The modified feedback error learning technique is used for on-line learning. Performance of the controller can be improved using an adequate target signal and reference trajectory. The proposed neural controller c...