Relay feedback process identification and controller design.
The aim of this thesis is to investigate relay feedback process identification and some controller
design methods. Using an exact analysis method, namely the state space method,
a set of conditions for the prediction of oscillations in relay control systems has been
developed. Since the exact solution for the limit cycle is found it becomes possible to
assess the stability of these oscillations using an eigenvalue criterion. A correction factor
has been introduced to overcome the limitations of the Balasubramanian's eigenvalue
criterion. Relay feedback identification in process control can lead to erroneous results
if the system parameters are estimated from the approximate describing function approach.
Exact analytical expressions are derived and on the basis of these expressions an
identification procedure is suggested which is capable of estimating the parameters of a
class of process transfer functions. Analytical expressions are presented for quantifying the
approximate estimation errors in the presence of measurement noise and load disturbance.
The performance limitations of the conventional PID controller have been clearly shown in
the context of controlling resonant, unstable or integrating processes. It has been shown
that a PI-PD controller with the PD in the inner loop not only avoids the derivative kick
but also a better performance is achieved than with a P or D controller in the inner loop.
Further, the same controller provides good disturbance rejection and its performance is
often near to that of an optimum controller for disturbance rejection and is significantly
better than the results of other design methods based on setpoint response. Simple tuning
methods based on standard forms for a PI-PD controller controlling time delay processes
have been presented which are particularly effective for integrating and unstable plants.
Automatic tuning formulae for PI, PID and PI-PD controllers have been proposed for
controlling stable and unstable processes.
The problem of controlling integrating and unstable processes incorporating time delay
has been tackled by proposing a new Smith predictor. It is shown that the predictor is
capable of successfully controlling stable, integrating and unstable processes. Controller
parameters leading to robust performance for various levels of uncertainty in the model
parameters particularly in the unstable time constant and time delay have been presented.
Also, simple and effective automatic tuning formulae are derived for the new Smith predictor
structure when the plant model is not available assuming first order model with
time delay for stable, unstable and integrating processes. The plant model blocks in the
control structure, as well as all the controllers, are designed from a single symmetrical