A characteristic of wood chip refiners is that the incremental gain be
tween the motor load and the plate gap is subject to a slow drift due
to plate wear and sudden changes in sign due to pulp pad collapse. A p
ad collapse can be caused by a change in operating point, or may occur
suddenly due to a feed rate or consistency disturbance. This poses a
problem for fixed-parameter linear controllers which may actually acce
lerate pad collapse and induce plate clashing as a result of getting c
aught in a positive feedback loop. The objective of this work is to de
velop a chip refiner motor load controller capable of detecting the on
set of pad collapse and, thereby, avoid plate clashing. The problem is
approached from a fault detection and control viewpoint where the ref
iner is treated as a linear dynamic system with fast time-varying para
meters. The proposed adaptive control algorithm consists of an improve
d parameter estimator and a controller containing ''dual'' features. T
o track both drifts and jumps in the parameters, a multi-model approac
h called adaptive forgetting through multiple models or AFMM is used.
A method of modifying the AFMM to include information about what to ex
pect in the event of a pad collapse is proposed. The main contribution
of the work is the development of an active adaptive controller or AA
C, which actively probes the system to improve the parameter estimates
. The use of active learning or probing in the controller is justified
by the fact that the parameter estimates are the key to identifying a
pad collapse, and that probing targets a portion of the input manipul
ations at continuously identifying these parameters. Finally, the AAC
is combined with the AFMM, and the resulting combination is tested on
an industrial refiner.