Multivariate Image Analysis in Mineral Processing - Advanced Control and Supervision of Mineral Processing Plants - page 124

0.6

N3

N4

0.4

N2

N5

0.2

N1

0

CL

-0.2

-0.4

BH

-0.6

-0.8

-0.4 -0.2 0 0.2 0.4 0.6

p 1

Figure 3.20 The p 1 N 5

correspond to the five bubble size signatures, ordered from smaller ( N 1 )tolarger( N 5 ). BH and CL
variables are the black holes and clear windows color features. From Liu et al. [37]

− p 2 loadings corresponding to the t 1 − t 2 score plot of Figure 3.19. N 1 −

In summary, froth health monitoring was investigated in this case study using
both wavelet bubble size signatures as well as some froth color features obtained us-
ing MR-MIA II. This approach was found to be very robust to noise and variations in
lighting conditions, and allows estimation of bubble size distribution without image
segmentation. Multivariate statistical process control charts could easily be estab-
lished in order to warn the operator that changes need to be made to reagent dosage
to maintain high metallurgical performance despite changes in ore feed properties.
These charts could be made available in the control room for all the flotation cells
or columns equipped with a digital camera. More appropriate and timely changes
to reagent dosage would result, as well as a more efficient use of operator work
time. An automatic froth health control scheme was also proposed by Liu et al.
[59] based on the monitoring strategy presented in this case study. It essentially
consists of computing the required control moves to implement reagent dosage in
order to bring froth health, quantified by PCA scores ( i.e. , t 1 t 2 in Figure 3.19),

back into a desired region. This control strategy uses both process data and images,
and accounts for process dynamics. The control scheme is cast into an optimization
problem. The reader is referred to Liu et al. [59] for more details on the approach.

−