Identification and control of nonlinear processes via reference system synthesis

This thesis presents research results towards the development of nonlinear control structures and the identification of nonlinear models from input/output process data. Reference System Synthesis as developed in the Process Modeling and Control Research Center at Lehigh University provided the vehicle for the study of design nonlinear controllers that will utilize such nonlinear input/output models as well as other nonlinear model forms. The development of Reference System Controllers for quasi-linear and nonlinear state-space models is presented. The method is shown to be unique in the resulting control structure, requiring direct feedback of the process output. The effect of high order reference systems on the performance of the corresponding controllers is also examined. It is shown that process induced dynamics appear in the closed-loop results when reference systems of order lower than the process are used.; A procedure for the identification of nonlinear processes is also presented. The bilinear model is shown to be able to represent process nonlinearities and is used as the basis of the identification algorithm. The representation of strictly observable state-space bilinear models by input/output recursive models is also presented. The parameters of the input/output model are identified from process data. Volterra series representation of the models provides constraints on the model parameters to ensure dynamic equivalence between the input/output and state-space representations. Balanced realization techniques are then used to calculate the bilinear state-space model matrices. Examples are given to show the effectiveness of the identification techniques. The examples highlight the use of bilinear models to represent nonlinear process behavior. The examples also demonstrate the effect of increasing the model dimension to better describe process nonlinearities.; The identification procedures have been coded as MATLAB programs in a Bilinear Identification Toolbox. The identification programs are described here. The techniques are used to identify nonlinear models for an industrially realistic process, a model IV fluid catalytic cracking unit simulation. The dynamic model used for the simulation was developed as part of this work and the model equations are presented. A bilinear model is identified for the carbon monoxide effluent of a fluid catalytic cracking unit in response to the lift air flow rate. The bilinear model is shown to outperform linear step response and parametric models.