Environmental Engineering Reference
In-Depth Information
respect to a given ore feed, as well as the opposite situation where under-dosage
leads to poor mineral recovery. Providing such information to operators in the con-
trol room could result in more stable operation, better mineral recovery and more
efficient use of operator's attention. Straightforward modifications to the proposed
approach could also yield froth grade predictions as discussed in the first case study.
The last case study presents the very difficult problem of estimating ROM ore
lithotype composition on conveyor belts based on dry and wet rock fragments of
highly heterogeneous visual appearance, but also showing significant similarities
between some rock types. This problem was selected to illustrate supervised clas-
sification of both spectral and textural features, requiring several steps due to the
level of complexity of the rock identification problem. On-line classification of rock
fragments according to their grade or grindability could be useful for modifying the
sorting/blending strategy before feeding AG/SAG grinding mills, for feedfoward
control of the mill's operating conditions (
e.g.
, throughput or water addition) and
for proactive adjustments to flotation reagent dosage. Very promising results were
obtained, particularly for dry rock mixtures. For wet rocks, specular reflection due
to wet surfaces and lower color contrast led to greater classification errors. Since
this was one of the first attempt to directly take into account surface moisture in the
rock type classification problem, further investigations on these issues are required.
In particular, adding imaging hardware to remove specular reflection combined with
contrast enhancement methods may help improve classification results for wet rock
mixtures. Overall, the results were found to be very promising and suggest future
work on this subject.
The multivariate imaging methods presented in this chapter have also been
used with great success in a wide variety of fields ranging from remote sensing
[25, 29, 95], combustion, pyrometallurgy, and the steel industry [34, 38, 39], for-
est products and pulp and paper [30, 31], snackfood industry [32, 33], plastics
[52, 96, 97], and even medical imaging [29, 98]. In most process cases, the vi-
sion sensors were used to develop new advanced control schemes that could not
be implemented using standard process instrumentation. The author hopes that this
chapter will help stir some interest in the mineral processing research community.
Fast, robust and reliable vision sensors are still required to improve mineral process
control, stability and productivity.
Acknowledgements
Financial support from the Natural Sciences and Engineering Research
Council of Canada (NSERC) and from COREM is greatly acknowledged. Thanks to the personnel
of Agnico-Eagle/Laronde and Xstrata Nickel Canada for providing the necessary materials and
access to their plants as well as for sharing their process knowledge. Special thanks go to Dr June
J. Liu and Professor John F. MacGregor for very insightful discussions on multivariate imaging
and for providing some of the materials used in this chapter. Finally, the author would like to thank
Gianni Bartolacci (COREM) for close collaboration in most of the case studies presented in this
chapter.
Search WWH ::
Custom Search