NONLINEAR CHEMICAL PROCESS CONTROL BY MEANS OF LINEARIZING STATE AND INPUT VARIABLE TRANSFORMATIONS

Difficult control problems in industrial chemical processes are usually characterized by multivariable, nonlinear behavior often accompanied by poor process input-output conditioning. An approach to dealing with the control of these types of systems is to construct control variable transformations that make the process appear to be less nonlinear and possibly less interactive. This work discusses the theoretical and practical limitations associated with several recently developed techniques for the implementation of this type of nonlinear multivariable control. In addition a new algorithm is proposed which greatly simplifies the generation of approximate linearizing transformations for high order multiple input nonlinear processes.;Central to the development of a set of linearizing transformations for a given process is the development of an accurate nonlinear dynamic model of the process. The development of full order and reduced order dynamic models of shell and tube heat exchangers and for high purity binary distillation columns is discussed in detail. Particular emphasis is placed upon the development of very low order, algebraically simple, nonlinear input-output models of high purity binary distillation columns. The use of the Singular Value Decomposition based process condition number was found to serve as a useful tool to guide selection of the process inputs to be used in the input-output model description. Proper selection of the process inputs was found to greatly improve the accuracy of the input-output model's simulation of the original process behavior.;An extension of the linear system Singular Value Decomposition based structural compensator was developed for use upon severely nonlinear systems. This form of structural compensator (decoupler) was shown to provide significantly improved performance over the conventional structural compensator designs when applied to a severely nonlinear ill-conditioned process. The use of this nonlinear structural compensator prior to the development of linearizing process transformations allows the multivariable linearization process to be treated as a family of single variable transformations thus greatly reducing the overall complexity of the transformation derivation.