System identification techniques for a rolling mill gauge control system

In industrial processes such as aluminum rolling mills, it is desirable for process improvement purposes, to develop a dynamic process model. Various methods have been considered to improve process performance without the use of such a model. Many of these methods are ad hoc and rely heavily on the process engineer's knowledge of the particular process.;Prior to this work, many of the models describing the rolling mill's thickness control loop behavior were either nonlinear, mechanistic models based on first principles, or linear empirical models.;Empirical modelling methods are critically evaluated in this thesis. These methods, being derived from process data, inherently match the process outputs. They predict the plant dynamic behavior and are of a simplistic form which allows or their use in real time control systems. The empirical techniques which are applied, include linear and nonlinear methods. The most common linear methods include the empirical transfer function estimate (ETFE), and the autoregressive with exogenous inputs (ARX) model. The parallel cascade method of nonlinear modelling is discussed in detail. (Abstract shortened by UMI.)...