1104 XXXI International Mineral Processing Congress 2024 Proceedings/Washington, DC/Sep 29–Oct 3
in a way that it is statistically representative (Zeiss 2007,
Fonseca 2011, Schmitt et al., 2016).
Other advanced methods of quantitative image analysis
based, for example, on Fourier descriptors (Bowman et al.,
2002, Sukumaran and Ashmawy 2001) can be also applied.
However, they can be useful only in basic and rather simple
investigations because a degree of complexity limits their
practical application to the particle shape characteristics.
To upgrade this approach, a methodology that identifies
the particle morphology should be used, which can gener-
ate relevant and easily interpretable descriptive parameters
(i.e., convexity, circularity).
Images of individual particles may bring information
not only on the particle’s size and shape, and other physical
properties of material, but may characterize the course of
crushing and grinding operations. Such information may
be used as a feedback for the process control, and optimiza-
tion, when, for example, a product with specific character-
istics is expected to be obtained. In the case of separation
operations, images of flotation froth may be used in collect-
ing information on flotation process characteristics. Such
visual information is difficult to obtain, but specific visual
systems are being implemented in the plant conditions, like
the FloVis system installed in KGHM, FrothSense+, cre-
ated by Metso Outotec and VisioFroth (Konieczny et al.,
2011). They can capture images of the froth from a flota-
tion machine and then they process it and calculate specific
froth parameters like size, shape and color of bubbles and
their mobility. They are used further to control the param-
eters of the flotation process.
Even though such systems help in flotation process con-
trol and are justified from the economic point of view, the
image parameters are not closely linked to the froth char-
acteristics, and knowledge about this relation could have
a great effect on improvement of effectiveness of flotation
process course and could improve the overall economy of
a processing plant. There are two main approaches toward
analyzing the images of the froth in the flotation process.
In the first one, the information is extracted directly from
the froth image (Konieczny et al., 2011) through the bub-
ble size, shape, velocity etc. The second approach is based
on the Fourier transform of the froth image (Sztaba and
Lenczowski 1993, Lenczowski et al., 1993). The image
parameters, called the descriptors, are linked to the froth
parameters, i.e., bubble size, shape, velocity, but these con-
nections are neither straightforward nor intuitive. These
parameters should be treated as statistical image descriptors
and are very useful in statistical image recognition processes.
The machine learning process (ML) provides the
ability to determine the relationship between the froth
characteristics and the froth image parameters. It is based
on the training groups of froth images, registered in various
stages of the flotation process. The ML process leads to the
Image Recognition (IR) algorithm which enables establish-
ing relationship between the froth image and froth char-
acteristics, understood as useful mineral content (Korder
and Lenczowski 1988, Galas et al., 1994, Galas 1994). To
perform that, however, the Image Analysis procedure must
be sensitive enough to detect significant variation in the
flotation process parameters. Interesting applications of IR
algorithms in the flotation technology can be found in the
literature (Galas et al., 1994, Lenczowski and Galas 1995,
Lenczowski and Galas 1998).
MATERIALS AND METHODS
The leading idea presented in this paper consists in showing
potential benefits of ore beneficiation in flotation opera-
tions. Application of vision systems to control the flotation
process course is becoming more popular. Specific sensors
of image transformation measure selected parameters of
flotation froth and on that basis the system can modify the
operational parameters of the process. The flotation process
will be carried out for sulphide copper ores and the entire
flotation consists of several operations including rougher,
main and cleaning flotation. The CCD camera observes the
surface of the flotation froth in selected flotation machine
(Figure 1). The image is transferred from the CCD cam-
era placed in the optical measurement point (OMP) to the
computer for further processing and analysis. The idea of
methodology can be characterized as follows:
a. The vision system captures images of the flotation
froth through a collection of cameras placed above
the surface of the flotation machine
b. Froth image processing algorithms determine the
values of parameters characterizing the flotation
froth. The parameters characterize structure, variabil-
ity, flow rate and transparency of the froth, as well as
the share of individual colors in the RGB and HSV
space.
c. The flotation process steering parameters are modi-
fied according to the obtained values of the visual
parameters.
The computer used in the experiments was a standard com-
mercial PC working under Windows. It was equipped with
the digital CCD camera, 512 × 512 pixels, 1 byte/pixel and
20 GB HDD for data. Authors of this paper have devel-
oped the specialized software for image acquisition and
processing. The Linear Discriminant Analysis was used for
correlation images with a Pb content in the flotation froth.
in a way that it is statistically representative (Zeiss 2007,
Fonseca 2011, Schmitt et al., 2016).
Other advanced methods of quantitative image analysis
based, for example, on Fourier descriptors (Bowman et al.,
2002, Sukumaran and Ashmawy 2001) can be also applied.
However, they can be useful only in basic and rather simple
investigations because a degree of complexity limits their
practical application to the particle shape characteristics.
To upgrade this approach, a methodology that identifies
the particle morphology should be used, which can gener-
ate relevant and easily interpretable descriptive parameters
(i.e., convexity, circularity).
Images of individual particles may bring information
not only on the particle’s size and shape, and other physical
properties of material, but may characterize the course of
crushing and grinding operations. Such information may
be used as a feedback for the process control, and optimiza-
tion, when, for example, a product with specific character-
istics is expected to be obtained. In the case of separation
operations, images of flotation froth may be used in collect-
ing information on flotation process characteristics. Such
visual information is difficult to obtain, but specific visual
systems are being implemented in the plant conditions, like
the FloVis system installed in KGHM, FrothSense+, cre-
ated by Metso Outotec and VisioFroth (Konieczny et al.,
2011). They can capture images of the froth from a flota-
tion machine and then they process it and calculate specific
froth parameters like size, shape and color of bubbles and
their mobility. They are used further to control the param-
eters of the flotation process.
Even though such systems help in flotation process con-
trol and are justified from the economic point of view, the
image parameters are not closely linked to the froth char-
acteristics, and knowledge about this relation could have
a great effect on improvement of effectiveness of flotation
process course and could improve the overall economy of
a processing plant. There are two main approaches toward
analyzing the images of the froth in the flotation process.
In the first one, the information is extracted directly from
the froth image (Konieczny et al., 2011) through the bub-
ble size, shape, velocity etc. The second approach is based
on the Fourier transform of the froth image (Sztaba and
Lenczowski 1993, Lenczowski et al., 1993). The image
parameters, called the descriptors, are linked to the froth
parameters, i.e., bubble size, shape, velocity, but these con-
nections are neither straightforward nor intuitive. These
parameters should be treated as statistical image descriptors
and are very useful in statistical image recognition processes.
The machine learning process (ML) provides the
ability to determine the relationship between the froth
characteristics and the froth image parameters. It is based
on the training groups of froth images, registered in various
stages of the flotation process. The ML process leads to the
Image Recognition (IR) algorithm which enables establish-
ing relationship between the froth image and froth char-
acteristics, understood as useful mineral content (Korder
and Lenczowski 1988, Galas et al., 1994, Galas 1994). To
perform that, however, the Image Analysis procedure must
be sensitive enough to detect significant variation in the
flotation process parameters. Interesting applications of IR
algorithms in the flotation technology can be found in the
literature (Galas et al., 1994, Lenczowski and Galas 1995,
Lenczowski and Galas 1998).
MATERIALS AND METHODS
The leading idea presented in this paper consists in showing
potential benefits of ore beneficiation in flotation opera-
tions. Application of vision systems to control the flotation
process course is becoming more popular. Specific sensors
of image transformation measure selected parameters of
flotation froth and on that basis the system can modify the
operational parameters of the process. The flotation process
will be carried out for sulphide copper ores and the entire
flotation consists of several operations including rougher,
main and cleaning flotation. The CCD camera observes the
surface of the flotation froth in selected flotation machine
(Figure 1). The image is transferred from the CCD cam-
era placed in the optical measurement point (OMP) to the
computer for further processing and analysis. The idea of
methodology can be characterized as follows:
a. The vision system captures images of the flotation
froth through a collection of cameras placed above
the surface of the flotation machine
b. Froth image processing algorithms determine the
values of parameters characterizing the flotation
froth. The parameters characterize structure, variabil-
ity, flow rate and transparency of the froth, as well as
the share of individual colors in the RGB and HSV
space.
c. The flotation process steering parameters are modi-
fied according to the obtained values of the visual
parameters.
The computer used in the experiments was a standard com-
mercial PC working under Windows. It was equipped with
the digital CCD camera, 512 × 512 pixels, 1 byte/pixel and
20 GB HDD for data. Authors of this paper have devel-
oped the specialized software for image acquisition and
processing. The Linear Discriminant Analysis was used for
correlation images with a Pb content in the flotation froth.