Soft Sensing - Advanced Control and Supervision of Mineral Processing Plants - page 144

application of soft sensors to this industry it is expected that gaps are filled in the
available reviews in papers and in topics on soft sensors.
In order to fulfill the objectives stated for a plant (circuit, system, unit, etc. )itis
essential to be able to extract information about the plant's condition, in order either
to drive it to a given operating point, or to a desired trajectory, or to such condition
that a performance measure is optimized. Such required information is obtained
directly or indirectly through sensors that measure plant variables. If any of these
measurements becomes unavailable, technical and economic objectives will not be
met because the plant's control, either automatic or manual, will be impaired and
the plant products will fail to meet specifications.
The sensor measurement may become unavailable because an installed sensor
system fails or its operation is interrupted for maintenance or repairs. A sensor for a
given measurement may also not be installed at all, e.g. , because it is not available
on the market, or due to the high cost of the sensor system and its maintenance.
A soft sensor may also provide measurement estimations when an actual sensor is
shared among several measuring points ( e.g. , X-ray copper grade analyzer) or, in
general, it can fill in measurements when sampling is infrequent.
An unmeasured variable, such as an unmeasured disturbance, prevents the system
from benefiting from information necessary for better plant performance. Through
changing an unmeasured disturbance into a measured disturbance, a sensor or a
soft sensor makes it possible to use such a measurement in the control of the plant
( e.g. , by means of feedforward control or other model-based control schemes). As a
result, it is sometimes possible to considerably improve the plant's operation. This is
particularly important when the unmeasured disturbance has a relatively large effect
on the controlled variables, as in some mineral processing plants.
On the contrary, when the sensor or soft sensor is not available, the use of ad-
vanced control strategies would become necessary. For example, if the controlled
variable is highly correlated with a random unmeasured disturbance, either some
form of stochastic control may be required, or a less efficient control system must
be used. But it is a well-known fact that complicated control strategies tend to be
vulnerable and of reduced robustness. As a result, it often happens that after several
problems come up, the control strategy is abandoned and control reverts to one that
is less elaborate, but surely less efficient. All this may be avoided if the sensor mea-
surement is available or if conditions are such that a soft sensor may be designed to
estimate the missing unmeasured disturbance.
Among the main unmeasured disturbances in mineral processing are lithology,
mineralization, liberation, grindability, and concentration of reactants in flotation
pulps. The absence of sensors for these plant variables implies that the soft sensor
must not only be designed using off-line measurement samples that are analyzed in
a laboratory, but also be updated using off-line samples.
There is clearly a question of economics in deciding whether a sensor's back up
should be to a spare sensor or to a soft sensor. Justification for a soft sensor depends,
on the one hand, on the relative cost of a real back-up sensor. For example, as a back
up for a wattmeter, a spare wattmeter may be best, which can easily be plugged in to
replace the missing one. On the other hand, for backing up other complex sensors,