| Summary: | Automatic tuning of the feedback controllers in a commercial paper machine CD control system
is considered in this thesis. A feedback controller tuning tool has been designed and developed for
tuning the Dahlin controller and control filter in the Honeywell-Measurex CD control system. This
tool includes the process identification module, the disturbance identification module and the
controller and filter tuning module. The tool is coded in MATLAB and some algorithms have been
implemented as "C" modules embedded into a product currently on the market.
The tuning strategy used in this tool is based on modern system identification and controller
parameter optimization techniques. A global search method combined with the Levenberg-
Marquardt refinement is used for identification of the paper machine time-response model (first
order plus delay). To obtain a longer disturbance realization sequence and increase data available
for identification of the process disturbance, CD residuals, extracted from bump test data, are used
for identifying an Integrated Moving Average disturbance model. The identification method is
based on the Recursive Extended Least Squares. Based on the identified process and disturbance
models, the Dahlin controller and control filter are tuned to minimize a LQG based performance
index. Since disturbance dynamics for a paper machine vary with time and with the operating
conditions, the possible changes of the disturbance characteristics should be taken into account in
the developed tuning algorithms. In the modified tuning scheme, the controller and filter are tuned
in such a way that a performance lower bound is guaranteed for all disturbances in a certain set.
The feedback controller tuning algorithms were tested extensively using simulation models,
Honeywell-Measurex Devron hardware-in-the-loop paper machine simulator and many sets of mill
data. A mill trial was carried out for validation of the disturbance identification and control loop variation prediction. Simulation and mill data tests showed that the developed identification
algorithms were applicable to real-life mill data and the process variations were predicted with
satisfactory accuracy. For better tractability of problem, in the developed algorithms it was
assumed that the disturbances in adjacent zones were independent and the actuator response was
narrow (about 1 or 2 zones wide). In most cases, the developed algorithms gave reasonable tuning
parameters as long as these assumptions held. Validation using mill data also showed that the
modified tuning algorithm sometimes overestimated the process and actuator variances and resulted
in too conservative a tuning. === Applied Science, Faculty of === Electrical and Computer Engineering, Department of === Graduate
|
|---|
