Sunteți pe pagina 1din 9

Minerals Engineering 98 (2016) 204–212

Contents lists available at ScienceDirect

Minerals Engineering
journal homepage: www.elsevier.com/locate/mineng

Statistical effect of sampling particle number on mineral liberation


assessment
Takao Ueda ⇑, Tatsuya Oki, Shigeki Koyanaka
Environmental Management Research Institute, Department of Energy and Environment, National Institute of Advanced Industrial Science and Technology (AIST), 16-1
Onogawa, Tsukuba, Ibaraki 305-8569, Japan

a r t i c l e i n f o a b s t r a c t

Article history: Appropriate assessment of mineral liberation is vital for the efficient operation of the entire mineral pro-
Received 31 May 2016 cessing process. Recently, the automated high speed analysis of ore particles that are mounted in resin,
Revised 17 August 2016 sectioned, and polished has become popular. However, these particle sectional analysis methods include
Accepted 26 August 2016
two inadequately studied errors: a statistical dispersion correlated with the number of analyzed particles
Available online 31 August 2016
and a stereological bias, which unavoidably overestimates the degree of liberation in 2D. In addition, even
if the stereological bias can be overcome using a correction method, the estimated degree of liberation in
Keywords:
3D also exhibits a statistical dispersion. A series of numerical analysis on more than one million biphase
Mineral liberation
Stereological bias
particles with various types of inter-particle structures was conducted to systematically investigate these
Statistical dispersion errors, and the previously proposed stereological correction method was attempted. A method was pro-
posed for estimating the number of particle sections that should be analyzed in order to achieve a degree
of apparent liberation in 2D and a degree of liberation in 3D for a desired arbitrary reliability. Validation
studies showed the reliability of the proposed method was averagely 97.6% in 2D and 3D.
Ó 2016 Elsevier Ltd. All rights reserved.

1. Introduction In addition, L2D obtained from the above-mentioned particle


sectional measurement process unavoidably includes a stereologi-
In order to achieve a high recovery rate, it is vital to liberate cal bias that overestimates the degree of liberation in 3D (L3D )
valuable minerals from gangue in the comminution process during (Gaudin, 1939). Stereological bias is caused by the fact that a sec-
mineral processing. Appropriate assessment of the degree of min- tion of a biphase particle has the possibility of being apparently lib-
eral liberation is also important when optimizing the entire crush- erated in 2D, depending on the particle’s internal structure and
ing and comminution process. sectional position, and a section of the liberated particle will inevi-
In general, comminuted ore particles are classified by size, tably be liberated in 2D as well. To overcome the stereological bias,
mounted in resin, sectioned and polished to assess the degree of several correction methods have been proposed, e.g., relatively
liberation. The sectional measurements can be traditionally con-
simple methods that estimates L3D using the quotient of L2D and
ducted using an optical microscope or with the recently popular-
empirical coefficients (Gaudin, 1939; Petruk, 1978) and methods
ized automated analyzer (e.g., Mineral Liberation Analyzer
that estimate the liberation distribution in 3D from that of sec-
(Fandrich et al., 2007), Quantitative Evaluation of Minerals by
tional areas (in 2D) or intersection lines (in 1D) using transforma-
Scanning Electron Microscopy (Gottlieb et al., 2000), and TESCAN
tion kernels (Gay and Morrison, 2006; King and Schneider, 1998;
Integrated Mineral Analyzer (TESCAN, 2012)). One of the main
Miller and Lin, 1988). These methods can successfully estimate
advantages of an automated analyzer over optical microscopy is
its high analysis speed (Sandmann, 2015). However, the effect of the L3D of well-examined samples, but their versatility has not been
the number of analyzed particle sections on the statistical disper- well studied. A correction method was developed using a com-
pletely different approach by evaluating the complexity of the tex-
sion of the measured degree of apparent liberation in 2D (L2D )
ture of the particle sections. The complexity of particle sectional
has not yet been systematically studied.
texture is first evaluated by the fractal dimension of the image
intensity (d) (Kaneko, 1987). A correlation between the stereologi-
cal bias and d and the areal fraction of one phase (F a ), which are
⇑ Corresponding author. measurable 2D parameters, is then prepared using an all-
E-mail address: t-ueda@aist.go.jp (T. Ueda). encompassing simulation of various phase patterns. Furthermore,

http://dx.doi.org/10.1016/j.mineng.2016.08.026
0892-6875/Ó 2016 Elsevier Ltd. All rights reserved.
T. Ueda et al. / Minerals Engineering 98 (2016) 204–212 205

the fraction of stereological bias is estimated only from the corre- In addition to the above mentioned process (Ueda et al., 2016b),
lation for the 2D parameters (d and F a ) (Ueda et al., 2016a). 250 patterns of [ii]-[iii] processes were conducted for a single con-
Even if these stereological correction methods successfully esti- dition of dA and F 0v to prepare an adequate number of particles to
mate L3D , the statistical dispersion originating from the number of investigate the statistical effect of the number of analyzed
analyzed particle sections should also be appropriately estimated. particles.
To the extent of the authors’ knowledge, few studies have been This procedure is interpreted as a modeling of the sample
conducted regarding the influence of the number of particle sec- preparation of non-preferential breakage of a biphase material
tion measurements on the statistical dispersion of the degree of and subsequent size classification with a relatively low computa-
liberation; one exceptional study was conducted by Leigh et al. tional load. It uses a similar approach to that of Gay (Gay, 2004,
(1997), who compared the stereological correction efficiency for 1999) in the sense that the particles are hollowed from biphase
2000 sectional samples with 250 sectional samples. materials; although, it is unique in the use of DEM for the random
In practice, proper sampling is important for correct assessment packing of particles.
of the parent population. According to Wills and Napier-Munn, the
most satisfactory sampling method is to gain the ore particles by 2.2. Analysis of particle structure
making a cut at right angles to the stream while its free fall dis-
charge (Wills and Napier-Munn, 2006). This study is based on Several types of volumetric and areal information were geomet-
the assumption that the sampling was appropriately conducted, rically calculated in 3D and 2D, respectively. In 3D, the particle vol-
and the sample is considered to be the representative of the parent ume (V) and the volume of the domains of phases A and B (V A ; V B )
population. Gy proposed an equation of the minimum weight of were calculated for each particle, where V ¼ V A þ V B . In particular,
sample required for general purpose with respect to the maximum when a particle is liberated with phases A or B, its particle volume
sample size and the variance of statistical distribution of the sam- is defined as V lib lib
A or V B , respectively.
pling values (Gy, 1979). In this study a model to determine mini- Similarly in 2D, the particle sectional area (S) and sectional
mum requirement of sample particle numbers, specifically, to areas of phases A and B domains (SA ; SB ) were calculated for each
measure the degree of mineral liberation will be considered. particle, where S ¼ SA þ SB . When a particle section is apparently
Based on the above-mentioned background, the goal of this liberated with phases A or B, its particle sectional area is defined
study is to establish a model to determine the required number
as Slib lib
A or SB , respectively.
of particle sectional measurements to obtain the degree of libera-
For sectional calculations, sample sections parallel to the bot-
tion in 2D and 3D with the arbitrary reliability and to validate
tom of the particle assembly were considered [Fig. 1(a)]. Seven sec-
the model by numerical simulation. The experimental validation
tions with a H of 2, 4, 6, 8, 10, 12, and 14 were calculated per
will be presented in a future study. The results of this study can
particle assembly. The difference between the sections was set as
be used for both efficient planning for the determination of the
2, the same as the maximum diameter of the particles, to avoid
required number of particle samples with an arbitrary acceptable
the double sectioning of particles. Since 250 patterns were pre-
error before analysis and for the estimation of the statistical error
pared, 1750ð¼ 7  250Þ sample sections were calculated; a total
included for an already measured degree of liberation from the
of 1,079,000 particle sections with random structures were
particle number after analysis.
obtained for a single condition of dA and F 0v .
On the other hand, for volumetric calculations, particles with a
height of the center point positioned between 2 and 14 were of
2. Methodology
concern; 5828 particles were analyzed per sample; a total of
1; 457; 000ð¼ 5828  250Þ particles were calculated for a single
2.1. Particle modeling
condition of dA and F 0v . The following parameters are defined as fol-
Particles were modeled using the discrete element method lows: areal fraction of phase A domain (F a ); degree of apparent lib-
 
(DEM) (Cundall and Strack, 1979) software ‘‘ESyS-Particle” eration in 2D for phases A and B L2D 2D
A ; LB ; volume fraction of
(Weatherley, n.d.). 7463 spherical particles with diameters ranging phase A domain (F v ); and degree of liberation in 3D for phases A
from 1.0 to 2.0 were generated at random positions in a cuboid  
with width W, depth D, and height H of 30, 30, and 20, respectively. and B L3D 3D
A ; LB .
Particles were allowed to fall freely in the cuboid until the total 2D parameters
kinetic energy of all the particles became negligibly small. Fig. 1 PN
(a) shows the packed particles. SA
F a ¼ P1N ; ð1Þ
The internal structures of biphase particles were modeled as a 1S
spherical core phase ‘‘phase A” dispersed in a matrix phase ‘‘phase
B”. Fig. 1(b) [i]-[iii] schematically illustrates the modeling proce- PN lib
1 SA
dure. Panel [i] depicts the particles after packing. As shown in L2D
A ¼ PN ; ð2Þ
panel [ii], mono-sized spherical phase A elements were generated 1 SA
at random positions in the cuboid. The diameter of phase A ele-
PN lib
ments (dA ) and the proportional volume of all the phase A elements
  1 SB
L2D
B ¼ PN ; ð3Þ
of the cuboid F 0v are given, and the number of phase A elements 1 SB

was calculated from dA and F 0v . Since the phase A elements were 3D parameters:
generated independently from the existence of the particles, they
PN
could be freely positioned inside or outside of the particles. In VA
F v ¼ P1N ; ð4Þ
panel [iii], the phase A domain of the particles is designated as
1V
‘‘phase A domain,” and the rest is designated as ‘‘phase B domain,”
respectively. For convenience, the initial state of phase A [ii] is PN lib
1 VA
called the ‘‘phase A element,” in contrast to the phase A domain L3D
A ¼ PN ; ð5Þ
in this study. 1 VA
206 T. Ueda et al. / Minerals Engineering 98 (2016) 204–212

Fig. 1. (a) Spherical particle assembly with positions of sample sections. (b) Conceptual diagrams of the biphase particle modeling procedure: [i] randomly generated
spherical particles are packed into a cuboid; [ii] phase A elements are generated at random positions in the cuboid; [iii] overlaps between particles and phase A elements were
set as the phase A domains in a particle, whereas the rest of the particle was set as phase B domains; 250 patterns of the [ii]-[iii] process was conducted to prepare an
adequate number of particles.

PN lib  
1 VB phase A elements F 0v , 0.05, 0.15, and, 0.25. Table 1 summarizes
L3D
B ¼ PN ; ð6Þ
1 VB the nine types with the number of phase A elements.
where N is the number of particle sections or particles. Fig. 2 depicts partial close-up sectional images of the nine par-
ticle types, which shows that a variety of particle textures were
F a ; L2D 2D 3D 3D
A ; LB ; F v ; LA , and LB take values ranging from zero to one
modeled using the above-mentioned procedure.
and from the principle of the stereological bias, L2D 3D
A P LA and
L2D 3D
B P LB , respectively. 4. Results
It can be noted that F v takes a larger value than F 0v . F 0v and F v are
volume fractions of phase A against the cuboid (Fig. 1(b) [ii]) and For particle sections in 2D, 20 groups were prepared in which N
that against the particles (Fig. 1(b) [iii]), respectively. The reason particle sections were randomly selected without duplication from
for this is assumed to be because when a phase A element is ran- the total of 1,079,000 particle sectional data, and the averages and
domly generated, the vicinity of the boundary of the cuboid standard deviations of F a , L2D 2D
A , and LB were calculated. 100 patterns
becomes sparser than the total volume fraction of a phase A ele-
of N ranging from 20 to 50,000 in equal steps were prepared in the
ment, and the same is true for particles. Therefore, the particles
log scale. It is noted that the particle sections were treated inde-
have a larger possibility of being imposed with a phase A element
pendently from the sample sections [Fig. 2].
than with F 0v .
For particles in 3D, F v , L3D 3D
A , and LB were calculated from the
For quantitative assessment of the effect of stereological bias on
total of 1,457,000 particles. Therefore, each of the nine particle
the degree of liberation, the difference between the degrees of
structure types have their own values of F v , L3D 3D
A , and LB ,
apparent liberation in 2D and that of liberation in 3D
  respectively.
L2D3D ; L2D3D , and the fraction of the overestimated degree of lib-
A B Table 2 tabulates L2D 2D 3D 3D
A ; LB ; LA ; and LB for all the nine types of
eration (rA , rB ) are defined as follows: internal particle structures. Here, L2D 2D
A and LB are the averages of

LA2D3D ¼ L2D  L3D 20 sample groups with N ¼ 50; 000.


A A ; ð7Þ
Fig. 3(a) and (b) shows the average values and standard devia-
tions of L2D 2D 3D 3D
A and LB for various N, together with LA and LB . In panel
LB2D3D ¼ L2D 3D
B  LB ; ð8Þ
(a), the standard deviation of L2D
A is large when N is small, and
decreases with increasing N. Even in the converged state with a
L2D3D
rA ¼ A
; ð9Þ
L2D
A
Table 1
Numerical conditions for nine types of particles.
L2D3D
rB ¼ B
; ð10Þ
L2D
B
Type Diameter of phase A Volume fraction of phase
 
Number of phase
element (dA ) A element F 0v A elements

where L2D3D
A , L2D3D
B , rA , and rB take values ranging from zero to 1 0.4 0.05 20,143
one. 2 0.4 0.15 60,429
3 0.4 0.25 100,715
4 2.0 0.05 161
3. Numerical simulations 5 2.0 0.15 483
6 2.0 0.25 806
Nine types of internal particle structures were prepared with 7 4.0 0.05 20
the combination of three values of the phase A element diameter 8 4.0 0.15 60
9 4.0 0.25 101
(dA ), 0.5, 2.0, and 4.0, and three values of the volume fraction of
T. Ueda et al. / Minerals Engineering 98 (2016) 204–212 207

Fig. 2. Partial close-up images of the sections of nine types of particle assemblies. The black circles denote particles, and gray and white regions in the particles represent
phase A and B domains, respectively.

Table 2
Calculated degrees of liberation for phases A and B in 2D and 3D.

Type Degree of liberation for phase A in 2D Degree of liberation for phase B in 2D Degree of liberation for phase A in 3D Degree of liberation for phase B in 3D
       
L2D
A L2D
B L3D
A L3D
B

1 0.000 0.410 0.000 0.062


2 0.000 0.061 0.000 0.000
3 0.000 0.009 0.000 0.000
4 0.113 0.881 0.038 0.769
5 0.111 0.635 0.038 0.386
6 0.110 0.401 0.037 0.148
7 0.412 0.948 0.275 0.906
8 0.408 0.830 0.276 0.703
9 0.405 0.689 0.277 0.492

considerably large N, L2D 3D


A exhibits a large difference from LA , which
In contrast with the true value of P, the sample value with N
is the true value in 3D. This difference is caused by the stereolog- b To estimate P from P
particle sections is defined as P. b within an
ical bias. The same is true of L2D
B in panel (b). Fig. 3(a) and (b) clearly
statistical error of n, the required number of particle sections
shows that the dispersion of the degree of apparent liberation in (N) is calculated as follows:
 
2D L2D 2D  2
A ; LB can be controlled by an appropriate number of parti-
KP b
NP b
Pð1  PÞ; ð12Þ
cle section analyses, but the effect of stereological bias cannot be n
diminished.
Statistical considerations were made to estimate the number of where K P is a value that satisfies the following equation, giving
particle sections required for measurements with an arbitrary sta- K P ¼ 1:96 for a reliability of 95% from the table of a normal
tistical dispersion. The following random variable (P) is defined for distribution:
the probabilistic approach:
1  0:95
8 P lib P ProbfK P 6 ug ¼ ; ð13Þ
>
> SA = S ¼ L2D
A Fa for phase A in 2D; 2
>
> P lib P
< SB = S ¼ L2D
B ð1  F a Þ for phase B in 2D; where u is a variant that obeys the standard normal
P ¼ P lib P ð11Þ distribution. According to Eqs. (11a)–(11d), n has the following
>
> V A = V ¼ L3D
>
> A Fv for phase A in 3D;
: P lib P relationship with an acceptable error for the degree of
V B = V ¼ L3D
B ð1  F v Þ for phase B in 3D: liberation (e):
208 T. Ueda et al. / Minerals Engineering 98 (2016) 204–212

 
Fig. 3. Average values and standard deviations of the degrees of apparent liberation in 2D L2D 2D
A ; LB for various number of particles (N) together with the degree of liberation
 
in 3D L3D 3D
A ; LB (a) for phase A and (b) for phase B.

8
>
>
eF a for phase A in 2D; b was determined as 0.030 by the calculation from
Here P
>
< eð1  F Þ for phase B in 2D; F a ¼ 0:27 and L2D 2D
A ¼ 0:11. When an acceptable error for LA (e) is
a
n¼ ð14Þ
>
> eF v for phase A in 3D; 0:05, n becomes 0:0135ð¼ e  F a ¼ 0:05  0:27Þ according to Eq.
>
:
eð1  F v Þ for phase B in 3D: (14a). As represented by the dashed-dotted line in Fig. 4,
n ¼ 0:0135 leads to N ¼ 602. Therefore, L2D
A can be obtained within
Fig. 4 shows the correlation among n, N, and Pb with the results
the acceptable error of 0:05 with a reliability of 95% from the
of Type 6 calculated using Eq. (12). The estimation method of the measurement of 602 particle sections.
required N from Fig. 4 will be explained using Type 6 of phase A Table 3 represents the number of particle sections required
as an example. According to Eq. (11a), L2D
A and F a are required to to estimate L2D and L2D within an e of 0:01, 0:02, and
A B
b Considering the objective of the estimation of the
determine P. 0:05. It is noted that the values of L2D
A for Types 1–3 and of
required number of particle sections to be measured, accurate val-
L2D
B for Type 3 became considerably small. This is because the
ues of L2D
A and F a are not necessary. Hence rough estimates of LA
2D
L2D 2D
A of Types 1–3 and LB of Type 3 are almost zero. Hence, even
and F a obtained from a preliminary test with a small number of
from the measurement of a small number of particle sections,
particle sections is applicable in practice. Note that in the case that
zero values of L2D and L2D can be estimated. However, since a
this method is applied in an apparatus, the following procedure is A B
value of N that is too small is not realistic in the sense that
applicable: N is assumed to be a tentative goal from L2D
A and F a
rough estimates of L2D and F a are required to obtain N from
obtained from a limited number of particle section measurements A
Eq. (12), N values smaller than 100 were modified to be 100.
(e.g., 100); an accurate N is achieved gradually as L2D
A and F a are For reference, original N values smaller than 100 before the
revised as the measurement proceeds.
modification are noted in brackets in Table 3. Table 3 shows
that the required number of particle section measurements
together with the calculated F a that corresponds to the mean
grade of the phase A in 2D. The required number of particles
dispersed greatly depending of the particle texture types. To
obtain a statistically reasonable measurement of L2D A and L2D
B ,
it is required to estimate an appropriate number of particle
sections using the proposed method.
The results of Table 3 were validated as follows:

(A) 54 cases (9 texture types with 3 values of e and 2 phase


types) with N particle sections listed in Table 3 were
considered.
(B) 100 samples of each case were prepared, and L2D 2D
A and LB
were calculated.
(C) L2D
A and L2D
B for the maximum Nð¼ 1; 079; 000Þ were
calculated and defined as L2D 2D
Amax and LBmax .
(D) The percentage of samples that satisfied
L2D 2D 2D
A  e 6 LAmax 6 LA þ e was calculated and denoted as
b (n) and number of particle
Fig. 4. Correlation between the statistical error of P the estimation reliability. The same was applied to L2D
B
sections (N) for an example with Type 6. as well.
T. Ueda et al. / Minerals Engineering 98 (2016) 204–212 209

Table 3
Number of particle sections required to estimate an apparent degree of liberation in 2D. The values in brackets are original values that were smaller than 100 and were modified
to be 100.

Type Fa Number of particle sections required to estimate degree of Number of particle sections required to estimate
   
liberation for phase A in 2D L2D
A degree of liberation for phase B in 2D L2D
B

e ¼ 0:05 0:02 0:01 e ¼ 0:05 0:02 0:01


1 0.05 100 (11) 100 (71) 286 407 2542 10,168
2 0.16 100 (4) 100 (24) 100 (97) 105 656 2622
3 0.26 100 (2) 100 (14) 100 (57) 100 (18) 111 443
4 0.05 3170 19,812 79,250 240 1498 5992
5 0.16 1028 6424 25,696 547 3417 13,670
6 0.27 602 3764 15,057 601 3755 15,021
7 0.05 11,439 71,492 285,968 160 998 3990
8 0.16 3603 22,519 90,076 464 2902 11,608
9 0.27 2026 12,662 50,647 727 4544 18,177

Table 4
Estimation reliability of the apparent degree of liberation in 2D.

Type Fa Estimation reliability of the degree of liberation for Estimation reliability of the degree of liberation for
phase A in 2D (%) phase B in 2D (%)
e ¼ 0:05 0:02 0:01 e ¼ 0:05 0:02 0:01
1 0.05 100 100 100 93 92 93
2 0.16 100 100 100 94 99 97
3 0.26 100 100 100 100 100 100
4 0.05 99 97 99 95 95 97
5 0.16 99 99 100 92 92 96
6 0.27 100 99 96 96 95 97
7 0.05 99 99 96 96 100 100
8 0.16 94 96 98 98 97 97
9 0.27 100 99 99 97 98 97

Table 4 shows the estimation reliability of the 54 cases obtained Types 1–3 became almost zero. Since the size of the phase A ele-
above. An average reliability of 97.61% was obtained, which vali- ment was less than the minimum diameter of the particles, both
dates the proposed estimation method of required number of par- L2D 3D
A and LA exhibit almost zero values for these types. As shown
ticle section measurements. in Fig. 5, the effect of stereological bias on the degree of liberation
In order to assess the magnitude of the stereological bias, the is large and greatly dispersed depending on the texture types of
difference between the degrees of apparent liberation in 2D and particle sections, which indicates the necessity of establishing an
 
that of liberation in 3D L2D3D ; L2D3D was calculated according appropriate stereological correction method.
A B

to Eqs. (7) and (8) with L2D 2D


A and LB for N ¼ 50; 000. Fig. 5 shows
L2D3D and L2D3D for all of the nine types of internal particle struc- 5. Discussion
A B
2D 3D
tures calculated from L and L values listed in Table 2. L2D3D
A of
A stereological correction method (Ueda et al., 2016a) was
developed in which the complexities of the texture of the particle
sections are assessed by a fractal dimension (d) of an imaginary
surface area of image intensity (Kaneko, 1987). Procedure to obtain
d is concisely explained as follows but more detail is in Ueda et al.
(2016b).

(A) On a cross-section of a sample, squares with a size of dmax


(maximum particle diameter) are superimposed on each
particle section.
(B) Each square is divided into N2 small squares with size r
(=dmax/N).
(C) On the small squares enveloped in a particle cross-section,
imaginary 3D structure with width and depth of r and with
height of image intensity is assumed and its surface area is
calculated.
(D) 50 cases of A(r) with N ranging from 1 to 50 are calculated. d
is obtained using Eq. (15) with a fitting line determined
using the least-squares method for A(r) plots on a double
logarithmic chart.

logAðrÞ ¼ ð2  dÞlogr þ C; ð15Þ


Fig. 5. Difference between the degrees of apparent liberation in 2D and that of
 
liberation in 3D L2D3D
A ; L2D3D
B for nine types of textures. where C is a constant.
210 T. Ueda et al. / Minerals Engineering 98 (2016) 204–212

A series of simulations on binary particles with various phase A Fig. 7(a) and (b) compares the average values and standard
0 0
element diameters (dA ) and volume fractions of phase A elements deviations of L2D 3D
and L2D , L3D
  A , LA B , together with the true degree
 B
F 0v were conducted in a similar fashion to our previous study of liberation in 3D LA ; LB . The estimated L3D
3D 3D 0
3D0
A and LB agree well
(Ueda et al., 2016a). 1470 types of binary particles with dA ranging with the true values of L3D 3D
A and LB , respectively, not only for Type 6,
from 0.2 to 5.0 in steps of 0.1 and with F 0v ranging from 0.01 to 0.3 but also for all the other types of samples. Since both the isogram
in steps of 0.01 were prepared, and the process shown in Fig. 1(b) and Types 1–9 were modeled using spherical particles and phase A
was conducted. elements, unsurprisingly, the stereological correction succeeded.
Fig. 6(a) and (b) shows isograms of the fraction of the overesti- The effectiveness of this stereological correction method on irreg-
mated degree of liberation (rA ; rB ) against F a and d, respectively, ular shaped particles has also been validated (Ueda et al., n.d.).
together with the results for Type 1–9 samples. These isograms 0 0
Here the statistical dispersion of L3D
A and L3D
B derived from L2D
A
enable us to obtain stereological bias information (rA ; rB ) from
and L2D
B was considered.
only easily obtainable 2D information (F a , d). For convenience, 0 0

the read values of rA and rB in Fig. 6 are redefined as r0A and r0B , Since L3D
A and L3D
B are multiplications of non-independent vari-
respectively. ables as shown in Eqs. (16) and (17), it has difficulty in calculating
 0  the variance directly from a probabilistic approach using Eqs.
3D0
The estimated degree of liberation in 3D L3D A ; LB was
(11c)–(11d), (14c)–(14d). Therefore, N was estimated using a frac-
obtained from r0A ; r0B , L2D 2D
A , and LB using the following equations: tion of the standard deviation of the degree of liberation in 3D and
0
0
in 2D (a) as follows:
L3D 2D
A ¼ ð1  rA ÞLA ; ð16Þ
 2
aK P b  PÞ;
b
0
0
NP Pð1 ð18Þ
L3D
B ¼ ð1  r 2D
B ÞLB : ð17Þ n

Fig. 6. Isograms of fractions of the overestimated degree of liberation (a) for phase A (rA ), and (b) for phase B (rB ) against the fractal dimension of an imaginary surface area of
image intensity d (Kaneko, 1987) and the area fraction of phase A (F a ). Unfilled circles show samples of Types 1–9.

   0 
3D0
Fig. 7. Average values and standard deviations of the degrees of apparent liberation in 2D L2D 2D
A ; LB and the estimated degree of liberation in 3D L3D
A ; LB , together with
 
the true degree of liberation in 3D L3D 3D
A ; LB .
T. Ueda et al. / Minerals Engineering 98 (2016) 204–212 211

where The combination of Fig. 8 and Eqs. (14a) and (14b), (18) enables
us to estimate the required number of particle section measure-
8P  
> 3D0 ments within an arbitrary acceptable error (e) of the degree of
>
> PN SD of LA2D for phase A;
< ðSD of LA Þ liberation.
N
a¼ P   ð19Þ Table 5 represents the required number of particle section mea-
>
> 0
>
: PN
SD of L3D
B
for phase B; surements N, together with the calculated F v that corresponds to
N
ðSD of LB Þ
2D
the mean grade of the phase A in 3D. The N values smaller than
100 for Types 1–3 were modified to 100 for the same reason as
where SD stands for the standard deviation. SDs were obtained with Table 3.
0 0
from 20 groups of L3D 3D 2D 2D
A , LB , LA , and LB with 100 patterns of N rang-
Table 6 shows the estimation accuracy of Table 5 validated
ing from 20 to 50,000 in equal steps in the log scale. using the same procedure as for Table 4. An average estimation
Fig. 8(a) and (b) shows isograms of a for phases A and B, respec- accuracy of 97.56% estimation was achieved using the proposed
tively, together with plots of Types 1–9 with their read a values. method.

Fig. 8. Isograms of the fraction of the standard deviation of the degree of liberation in 3D and in 2D (a) (a) for phase A, and (b) for phase B against the fractal dimension of an
imaginary surface area of image intensity (d) (Kaneko, 1987) and the area fraction of phase A (F a ). Unfilled circles show samples of Types 1–9.

Table 5
Number of particle sections required to estimate the degree of liberation in 3D. The values in brackets are original values that were smaller than 100 and were modified to be 100.

Type Fv Number of particle sections required to estimate the Number of particle sections required to estimate the
 0  0
degree of liberation for phase A in 3D L3D
A degree of liberation for phase B in 3D L3D
B

e ¼ 0:05 0:02 0:01 e ¼ 0:05 0:02 0:01


1 0.05 100 (0) 100 (0) 100 (0) 100 (92) 575 2301
2 0.16 100 (0) 100 (0) 100 (0) 100 (0) 100 (0) 100 (0)
3 0.26 100 (0) 100 (0) 100 (0) 100 (0) 100 (0) 100 (0)
4 0.06 815 5094 20,375 873 5449 21,797
5 0.17 185 1154 4615 883 5517 22,071
6 0.27 104 648 2591 335 2093 8373
7 0.06 8540 53,376 213,504 676 4219 16,867
8 0.18 2914 18,210 72,840 1592 9955 39,820
9 0.29 1596 9974 39,897 1581 9883 39,536

Table 6
Estimation reliability of the degree of liberation in 3D.

Type Fv Estimation reliability of the degree of liberation for Estimation reliability of the degree of liberation for
 0  0
phase A in 3D L3DA (%) phase B in 3D L3DB (%)

e ¼ 0:05 0:02 0:01 e ¼ 0:05 0:02 0:01


1 0.05 100 100 100 93 97 97
2 0.16 100 100 100 100 100 100
3 0.26 100 100 100 100 100 100
4 0.06 97 96 100 96 98 96
5 0.17 96 98 94 93 92 92
6 0.27 98 99 99 94 94 94
7 0.06 97 98 99 99 97 98
8 0.18 100 96 96 95 98 99
9 0.29 97 97 99 94 98 98
212 T. Ueda et al. / Minerals Engineering 98 (2016) 204–212

6. Conclusion Gaudin, A.M., 1939. Principles of Mineral Dressing. McGraw-Hill INC., ISBN-10:
0070230307, ISBN-13: 978–0070230309.
Gay, S.L., 2004. Simple texture-based liberation modelling of ores. Miner. Eng. 17,
A series of numerical simulations on the degree of liberation of 1209–1216. http://dx.doi.org/10.1016/j.mineng.2004.06.032.
various types of biphase particles was conducted. Using a previ- Gay, S.L., 1999. Numerical verification of a non-preferential-breakage liberation
model. Int. J. Miner. Process. 57, 125–134. http://dx.doi.org/10.1016/S0301-
ously proposed stereological correction method, the statistical dis-
7516(99)00011-3.
persion caused by the number of particle section measurements of Gay, S.L., Morrison, R.D., 2006. Using two dimensional sectional distributions to
the apparent degree of liberation in 2D (L2D ) and that of the esti- infer three dimensional volumetric distributions - validation using tomography.
0 Part. Part. Syst. Charact. 23, 246–253.
mated degree of liberation in 3D (L3D ) was systematically studied. Gottlieb, P., Wilkie, G., Sutherland, D., Ho-Tun, E., Suthers, S., Perera, K., Jenkins, B.,
The following conclusions were derived from the results and Spencer, S., Butcher, A., Rayner, J., 2000. Using quantitative electron microscopy
for process mineralogy applications. J. Miner. Met. Mater. Soc. 52, 24–25. http://
discussion. dx.doi.org/10.1007/s11837-000-0126-9.
Gy, P.M., 1979. Sampling of Particulate Materials Theory and Practice. Elsevier
0
Scientific Publishing Company.
1. Statistical dispersions of L2D and L3D decrease with an increase
Kaneko, H., 1987. Fractal feature and texture analysis. IEICE Trans. Inf. Syst., 964–
in the number of particle sections measured. 972, Japanese Ed J70-D.
2. An estimation method for the required number of particle sec- King, R.P., Schneider, C.L., 1998. Stereological correction of linear grade distributions
for mineral liberation. Powder Technol. 98, 21–37. http://dx.doi.org/10.1016/
tions to obtain L2D with a desired arbitrary acceptable error was S0032-5910(98)00013-8.
established using a probabilistic approach, and exhibited an Leigh, G.M., Lyman, G.J., Gay, S.L., 1997. Computer-based examination of self-
average estimation accuracy of 97.61%. contained, flexible stereological correction procedures. Powder Technol. 92,
101–110. http://dx.doi.org/10.1016/S0032-5910(97)03220-8.
3. An estimation method for the required number of particle sec- Miller, J.D., Lin, C.L., 1988. Treatment of polished section data for detailed liberation
0
tions to obtain L3D with a desired arbitrary acceptable error was analysis. Int. J. Miner. Process. 22, 41–58.
Petruk, W.B., 1978. Correlation between grain sizes in polished section with sieving
established by modifying the above-mentioned estimation for
data and investigation of mineral liberation measurements from polished
L2D with the dataset of a fraction of the statistical dispersion sections. Trans. Inst. Min. Metall. Sec. C 87, C272–C277.
0 Sandmann, D., 2015. Method Development in Automated Mineralogy. TU
between L3D and L2D and exhibited an average estimation accu- Bergakademie Freiberg.
racy of 97.56%. TESCAN, 2012. TESCAN Introduces the TIMA Mineralogy Solution. [WWW
Document]. <http://www.tescan.com/en/news/tescan-introduces-tima-
mineralogy-solution> (Accessed 7.22.16).
These results can be used for both efficient planning of the
Ueda, T., Oki, T., Koyanaka, S., 2016a. Stereological bias for spherical particles with
required number of particle samples with a desired arbitrary various particle compositions. Adv. Powder Technol. http://dx.doi.org/10.1016/
acceptable error before analysis and the estimation of the statisti- j.apt.2016.06.016.
Ueda, T., Oki, T., Koyanaka, S., 2016b. Numerical simulations of stereological bias in
cal error included for an already measured degree of liberation
particles with simple texture. Powder Technol. 298, 130–136. http://dx.doi.org/
from the particle number after analysis. 10.1016/j.powtec.2016.05.025.
Ueda, T., Oki, T., Koyanaka, S., n.d. Effect of particle shape on the stereological bias of
References the degree of liberation of biphase particle systems. Mater. Trans. (under rev).
Weatherley, D., n.d. ESyS-particle: HPC discrete element modelling software
[WWW Document]. <https://launchpad.net/esys-particle> (Accessed 7.22.16).
Cundall, P.A., Strack, O.D.L., 1979. A discrete numerical model for granular Wills, B.A., Napier-Munn, T.J., 2006. Wills’ Mineral Processing Technology: An
assemblies. Geotechnique 29, 47–65. Introduction to the Practical Aspects of Ore Treatment and Mineral Recovery.
Fandrich, R., Gu, Y., Burrows, D., Moeller, K., 2007. Modern SEM-based mineral Butterworth-Heinemann.
liberation analysis. Int. J. Miner. Process. 84, 310–320. http://dx.doi.org/
10.1016/j.minpro.2006.07.018.

S-ar putea să vă placă și