| dc.description.abstract | An adaptive optimizer is designed for control of multi-variable processes where the
operating characteristics of the plant are time varying. Based on the original sequential
method for static applications, the modified direct search method is made adaptive by
allowing only limited expansion and contraction and by re-measuring all points of a
simplex when improvement of the simplex position is not possible by reflection of the
vertices, and when expansion or contraction occurs. In one method, suitable for
performance functions that have a known optimum value, i.e., sums of squares of errors
from target values, expansion and contraction is determined by a threshold value of the
performance function. In a second method, expansion and contraction is determined by
ranking the newest measurement against the historical measurement values in the
simplex, similar to the Nelder-Mead method.
The optimizer was tested on simulated two by two multi-input multi-output first order
plus delay processes. It worked best in the absence of dynamics, but could give
acceptable performance where the delay and time constants were less than five
measurement sample times.
The adaptive optimizer is applied also to a simulated model of a thermo-mechanical
pulping screening room to control the accepts fibre distribution, using the main line and
reject screen volumetric rejects ratios. Two types of screens are modeled: screens with
smooth holed apertures in the screening baskets, and screens with slotted apertures. The
adaptive optimizer is capable of controlling the accepts fibre distribution over limited
ranges of disturbances in mean fibre length. The control range o f the slotted screen is
narrower than that of the holed screen because of its flatter fractionation profile.
Transport and capacity lags do not have a great impact because most of the fibre flow
follows the main accepts line. | |
