XXXI International Mineral Processing Congress 2024 Proceedings/Washington, DC/Sep 29–Oct 3 3749
and with the liner/lifters (Watson and Morrison, 1986
Watson and Morrison, 1985).
In a similar study using a purpose-built laboratory-
scale AG/SAG mill, model quartz (hard mineral) produced
the most noise intensity (RMS, PSDE and DWT feature
extractions), followed by iron ore (intermediate hard min-
eral) and model calcite (soft mineral) within 5 minutes
of grinding time, reflecting their different hardness prop-
erties. Beyond the 5 minutes of grinding time, the maxi-
mum acoustic emission of the grinding characteristics of
the minerals was shifted to model calcite, followed by iron
ore and model quartz (Owusu et al., 2020c Owusu et al.,
2022b). In general, the acoustic emission of the three min-
erals increased continuously with increasing time as the
particle size reduced (as illustrated in Figure 6 with model
quartz and iron ore), which contradicts the results obtained
from the laboratory ball mill. The resulting pattern corre-
sponded to the findings reported in the literature (Zeng
and Forssberg, 1994). This can also be explained by the
predominant grinding mechanisms associated with the lab-
oratory ball mill and AG/SAG mill. The differences in the
mechanisms of the two mills are influenced mainly by aspect
ratio (diameter to length ratio), along with other factors
like rotational speed, ball/rock ratio and lifter height and
configuration. In ball mills, grinding is primarily achieved
by abrasion milling whereas impact milling is the norm for
AG/SAG mills. The increasing mill noise in the AG/SAG
mill suggests that, as the ore particles break down, their
collision-blocking abilities are also reduced and unable to
absorb the impact energy of the cataracting steel balls. In
the ball mill, on the other hand, the ore particle size distri-
bution and the slurry coating can resist the interaction of
balls and liner/lifter from the dominant abrasion milling.
Nonetheless, different ore hardness can be distinguished in
grinding mills by tracking and analysing their mill acoustic
emission responses.
In another experiment, the effect of different feed
ore size distributions on acoustic emissions was carried
out under the same grinding environment (Owusu et al.,
2021g). It was evident from the nine feed size classes inves-
tigated (–2 +0.85 mm, –4 +2 mm, –6.7 +4 mm, –8 +
6.7 mm, –9.5 +8 mm, –13.2 +9.5 mm, 16 +13.2 mm,
–19+16 mm and –26.5+19 mm) that increasing the feed
size distribution increased the mill noise intensity, which is
in good agreement with the study by Das et al. 2011, where
vibration sensing was used. This result pattern is generally
demonstrated in Figure 7.
The coarser the feed sizes, the greater the mass or
weight of the discrete feed size which lower slurry viscosity
and contributes to increased mill noise, and vice versa. The
binary mixture of relative coarser feed size (–13.2 +9.5 mm)
and finer feed size (–4 +2 mm) fraction in ratios of 1:3, 1:1,
and 3:1 reduced the mill acoustic as the concentration of
the finer feed fraction was increased. This means that the
concentration of either coarse or fine feed size distribution
in the binary mixture could affect the grinding dynamics in
terms of PSD and acoustic emissions. Employing six dif-
ferent machine learning classification algorithms, the SVM
(support vector machine), LDA (linear discriminant analy-
sis), and ensemble algorithms emerged as the most suitable
algorithms that can be combined with features extracted
from AG/SAG mill acoustic to classify the feed size dis-
tribution (detailed of the study can be found in Owusu
et al., 2023). The acoustic feature extraction techniques
included the PSDE (power spectral density estimate),
DWT (discrete wavelet transform), WPT (wavelet packet
Figure 4. (A) Acoustic response of different lifter heights and
(B) PSD of different lifter heights, during milling (Owusu et
al., 2021a Owusu et al., 2021h)
and with the liner/lifters (Watson and Morrison, 1986
Watson and Morrison, 1985).
In a similar study using a purpose-built laboratory-
scale AG/SAG mill, model quartz (hard mineral) produced
the most noise intensity (RMS, PSDE and DWT feature
extractions), followed by iron ore (intermediate hard min-
eral) and model calcite (soft mineral) within 5 minutes
of grinding time, reflecting their different hardness prop-
erties. Beyond the 5 minutes of grinding time, the maxi-
mum acoustic emission of the grinding characteristics of
the minerals was shifted to model calcite, followed by iron
ore and model quartz (Owusu et al., 2020c Owusu et al.,
2022b). In general, the acoustic emission of the three min-
erals increased continuously with increasing time as the
particle size reduced (as illustrated in Figure 6 with model
quartz and iron ore), which contradicts the results obtained
from the laboratory ball mill. The resulting pattern corre-
sponded to the findings reported in the literature (Zeng
and Forssberg, 1994). This can also be explained by the
predominant grinding mechanisms associated with the lab-
oratory ball mill and AG/SAG mill. The differences in the
mechanisms of the two mills are influenced mainly by aspect
ratio (diameter to length ratio), along with other factors
like rotational speed, ball/rock ratio and lifter height and
configuration. In ball mills, grinding is primarily achieved
by abrasion milling whereas impact milling is the norm for
AG/SAG mills. The increasing mill noise in the AG/SAG
mill suggests that, as the ore particles break down, their
collision-blocking abilities are also reduced and unable to
absorb the impact energy of the cataracting steel balls. In
the ball mill, on the other hand, the ore particle size distri-
bution and the slurry coating can resist the interaction of
balls and liner/lifter from the dominant abrasion milling.
Nonetheless, different ore hardness can be distinguished in
grinding mills by tracking and analysing their mill acoustic
emission responses.
In another experiment, the effect of different feed
ore size distributions on acoustic emissions was carried
out under the same grinding environment (Owusu et al.,
2021g). It was evident from the nine feed size classes inves-
tigated (–2 +0.85 mm, –4 +2 mm, –6.7 +4 mm, –8 +
6.7 mm, –9.5 +8 mm, –13.2 +9.5 mm, 16 +13.2 mm,
–19+16 mm and –26.5+19 mm) that increasing the feed
size distribution increased the mill noise intensity, which is
in good agreement with the study by Das et al. 2011, where
vibration sensing was used. This result pattern is generally
demonstrated in Figure 7.
The coarser the feed sizes, the greater the mass or
weight of the discrete feed size which lower slurry viscosity
and contributes to increased mill noise, and vice versa. The
binary mixture of relative coarser feed size (–13.2 +9.5 mm)
and finer feed size (–4 +2 mm) fraction in ratios of 1:3, 1:1,
and 3:1 reduced the mill acoustic as the concentration of
the finer feed fraction was increased. This means that the
concentration of either coarse or fine feed size distribution
in the binary mixture could affect the grinding dynamics in
terms of PSD and acoustic emissions. Employing six dif-
ferent machine learning classification algorithms, the SVM
(support vector machine), LDA (linear discriminant analy-
sis), and ensemble algorithms emerged as the most suitable
algorithms that can be combined with features extracted
from AG/SAG mill acoustic to classify the feed size dis-
tribution (detailed of the study can be found in Owusu
et al., 2023). The acoustic feature extraction techniques
included the PSDE (power spectral density estimate),
DWT (discrete wavelet transform), WPT (wavelet packet
Figure 4. (A) Acoustic response of different lifter heights and
(B) PSD of different lifter heights, during milling (Owusu et
al., 2021a Owusu et al., 2021h)