SYSTEM IDENTIFICATION AND MODEL PREDICTIVE CONTROL FOR INTERACTING SERIES PROCESS WITH NONLINEAR DYNAMICS

This thesis discusses the empirical modeling using system identification technique and the implementation of a linear model predictive control with focus on interacting series processes. In general, a structure involving a series of systems occurs often in process plants that include processing sequences such as feed heat exchanger, chemical reactor, product cooling, and product separation. The study is carried out by experimental works using the gaseous pilot plant as the process. The gaseous pilot plant exhibits the typical dynamic of an interacting series process, where the strong interaction between upstream and downstream properties occurs in both ways.
The subspace system identification method is used to estimate the linear model parameters. The developed model is designed to be robust against plant nonlinearities. The plant dynamics is first derived from mass and momentum balances of an ideal gas. To provide good estimations, two kinds of input signals are considered, and three methods are taken into account to determine the model order. Two model structures are examined. The model validation is conducted in open-loop and in closed-loop control system.
Real-time implementation of a linear model predictive control is also studied. Rapid prototyping of such controller is developed using the available equipments and software tools. The study includes the tuning of the controller in a heuristic way and the strategy to combine two kinds of control algorithm in the control system.
A simple set of guidelines for tuning the model predictive controller is proposed. Several important issues in the identification process and real-time implementation of model predictive control algorithm are also discussed. The proposed method has been successfully demonstrated on a pilot plant and a number of key results obtained in the development process are presented.