Soft Sensing - Advanced Control and Supervision of Mineral Processing Plants - page 147

simulated plants. Examples of soft sensor design for industrial ore concentrators are
giveninSection4.4.
In the future, we expect that the developments of soft sensors will be driven by
the development of new data processing and sensor technologies in general, and of
new equipment in the mineral processing industry.
Nanotechnologies will permit the development of networks of a great number
of miniature sensors profusely located almost anywhere in the processing units and
connected to many miniature high-speed processors through newly developing wire-
less communications systems. Paralleling these developments, these new sensors
may have their corresponding soft sensors based on these nanotechnologies. Also,
the development of soft sensors for which no sensors have been developed may also
be favored. It is not difficult to imagine also that new data processing and communi-
cations technologies may also have a favorable impact on soft sensor shell systems
[5].
Thus, the new technologies point to an omnipresence of multiple sensors and soft
sensors in the processing units together with the pervasiveness of miniature signal
processing computers.
It is also reasonable to expect that the development of new equipment based on
new technologies may require new sensors and soft sensors. In order to improve
the efficiency of the comminution processes new equipment is being introduced in
the mineral processing industry, e.g. , the HPGR mill (high pressure grinding rolls
mill) [6] and the IsaMill [7]. Concerning flotation, a potentially increasing use of
the Jameson cell is also to be mentioned, where this flotation technology produces
an intense flotation recovery process using robust equipment [8].

p

v

y

PLANT

u

y

η

SOFT SENSOR
SYSTEM

Figure 4.1 Soft sensor system output y replacing failed sensor signal y using a model having
as inputs correlated secondary measurements: commands u , measured disturbances v , and other
plant outputs η. Although the unmeasured disturbances p affect the plant output y , they cannot be
directly used as inputs to the model