Abstract
M.Ing. (Mechanical Engineering)
This dissertation describes the performance of an Adaptive
controller and a Neural Network controller for water level
control in a steam drum, and compares them to the performance of
a conventional PIO controller in the same application.
The control problem is in essence the regulation of the speed of
a boiler feed pump in order to maintain a constant level in the
drum of a small model of a boiler. As a starting point, the
hydraulics and dynamics of the system are analyzed and the system
is shown to be nonlinear. A nonlinear computer simulation is
created and its response is compared with that of the real plant.
The simulation proves to be a close representation of the real
plant and it is used as an aid in creating a linear mathematical
model.
A set of control specifications are drawn up and a PIO controller
is designed for the plant. With the aid of a root locus diagram
it is shown that the plant cannot be controlled within specifications
under PID control. This shortcoming is then demonstrated
on both the linear mathematical model and the nonlinear plant.
Consequently, advanced control techniques are investigated in an
attempt to control the plant within specifications.
Different methods of adaptive control are discussed and a direct
model reference adaptive controller is designed. The least
squares algorithm for parameter adjustment is discarded in favour
of the slower gradient algorithm when it becomes apparent that
the wave motion inside the drum has an adverse effect on the
former algorithm. The control results obtained with both the
linear model and the real plant proves adaptive control to be
superior to PIO control in this application.
Additionally, the application of neural networks in control
systems is discussed. An adaptive neural network controller is
designed but is discarded due to instability caused by imperfect
modelling of the system...