Environmental Engineering Reference
In-Depth Information
(
i.e.
, create over- and under-dosage for a given ore feed). These tests were also per-
formed on the zinc flotation circuit (contact cell) of Agnico-Eagle/Laronde's plant
(Figure 3.13).
Two plant tests were made on two different days 7 weeks apart. These involve
steps changes in the activator (CuSO
4
) and collector (KAX) flow rates. These step
changes were maintained long enough to reach steady-state while all other oper-
ating conditions were maintained constant. A total of 487 and 456 images were
collected during the activator and collector tests, respectively. The digital camera
provided 24 bit RGB images with a spatial resolution of 720
480 pixels. Images
were captured every minute. For each image, a feature vector of 1
×
×
7 is computed
using the methods described above (
i.e.
, five size signatures,
N
1
N
5
,BHandCL).
After merging the feature vectors of all images captured in the two tests, the feature
matrix
X
F
−
7) was obtained. This data contains six different steady-states as
well as the transitions between them. Principal component analysis was then applied
to
X
F
to verify whether froth health could be tracked using a simple unsupervised
classification of the froth visual characteristics. The steady-states were found easily
distinguished in the
t
1
(
943
×
t
2
score plot shown in Figure 3.19.
The PCA score space is clearly able to track the evolution of froth health, from a
poorly loaded froth (
e.g.
, image 840), characterized by large bubbles and the pres-
ence of clear windows, to a very heavily loaded froth (
e.g.
, image 413). The latter
almost look like oatmeal and BH appear on the surface. Froth collapse could occur
if no control actions are made to improve this condition (
i.e.
, reduce collector and/or
activator dosage). Optimal operating conditions would fall somewhere in between
these two extremes. Since the steady-states are clearly distinguished in between
these extreme conditions (from left to right in the score plot), it would be straight-
forward to develop a multivariate statistical process control chart to warn operators
that suboptimal conditions are detected and control actions are required.
The movements from left to right in the score space can be interpreted in terms
of changes in the froth characteristics using the corresponding
p
1
−
p
2
loading plot
shown in Figure 3.20. The froth images having high CL,
N
4
and
N
5
values (but
lower values for the other variables) will fall on the left hand side on the score plot.
Indeed, their
p
1
loadings are negative, and since the scores are linear combinations
of all variables, negative
t
1
values will be obtained. Images having large amounts of
CL and larger bubble size (higher values of
N
4
and
N
5
) are associated with poorly
loaded froth (see image 840 in Figure 3.19). On the other hand, the heavily loaded
froths falling on the right hand side in the score plot (
i.e.
, positive
t
1
values in Figure
3.19), such as image 19 and 413, have smaller bubble sizes (
i.e.
, higher values of
N
1
,
N
2
,and
N
3
) and more BH. This is shown in Figure 3.20 where the
p
1
loadings for
these variables are positive (
i.e.
, high values for these variables will push
t
1
values
in the positive quadrant). The excursion in the negative
t
2
values (right hand side of
the score plot) are associated with the presence or absence of BH on heavily loaded
froths. For example, images 19 and 413 have similar bubble sizes (
i.e.
,theyhave
similar
t
1
values) but image 413 has some BH. The higher value for BH brings the
t
2
score values into the bottom right quadrant. The appearance of BH indicates the
onset of froth collapse.
−
Search WWH ::
Custom Search