University
stringclasses 19
values | Text
stringlengths 458
20.7k
|
|---|---|
Colorado School of Mines
|
Hardness values presented in Table 4.8 for the samples heated to 875 °C for 1200 s indicated, on
average, higher and more consistent hardness compared to those from HT1. These hardness data suggest
that all or most of the cementite was able to dissolve during the longer austenitization step, leading to a
more C-rich matrix and an improvement in mechanical properties. As retained austenite content was
consistently not affected by Nb content or quenching parameters, XRD was not performed on the
HT2-WQ or HT2-150Q samples.
XRD again did not reveal a trend between Nb content and volume fraction of retained austenite.
Additionally, there was no consistency between quench parameters and an increase in retained austenite
content. Rather, XRD results were similar between the HT1-WQ and HT1-150Q samples for each alloy.
As observed in Table 5.6, M is expected to increase by 5 - 10 ºC with each increment in Nb content. The
s
Koistinen-Marburger relationship in Equation 3.8 illustrates how an increase in M temperature
s
corresponds to a decrease in expected retained austenite content when quenched to the same temperature.
Thus, Equation 3.8 was employed to predict the retained austenite content of each alloy in conjunction
with experimentally determined M temperatures in Table 5.6 for a quench temperature of 150 ºC,
s
compared to the direct water quench to room temperature. Calculated retained austenite values are
included in Table 5.7.
Table 5.7 Predicted Retained Austenite Content Following 150 ºC Quench
Retained Austenite Content (vol pct)
Nb (wt pct)
Predicted Measured
0.01 66.8 22.1
0.25 59.2 21.0
0.5 55.6 22.5
1.0 48.2 19.4
These values differ considerably from measurements determined via XRD. The decreased
retained austenite content measured experimentally may be due to a high instability of austenite, possibly
stemming from C remaining tied up in Fe carbides, removing C from the austenite. Additionally,
Equation 3.8 only predicts the untransformed austenite content at 150 °C and much of that austenite is
expected to transform upon later cooling to room temperature. However, during the slow air quench to
room temperature following the 150 ºC quench, some C may partition from the martensite to the
austenite, increasing the austenite stability as time is increased. However, this effect is likely minimal as
these steels were not designed to have a significant response to partitioning.
101
|
Colorado School of Mines
|
5.7 Wear Testing
This section discusses results obtained from Bond abrasion and dry sand/rubber wheel (DSRW)
wear testing. Specific topics discussed include starting microstructure, mass loss trends as a function of
Nb content, deformed microstructures, and abrasive wear mechanisms.
5.7.1 Bond Abrasion Wear Data
Bond abrasion testing combines both abrasion and impact wear mechanisms for a complex wear
response. Increasing Nb content was not observed to increase Bond wear resistance as shown in
Figures 4.35 and 4.41. While Figure 4.34 did show favorable results in that wear rate dropped drastically
once 0.25 wt pct Nb was added to the steel, this behavior can be understood in the context of the
decarburized surface layer resulting from the heat treatment, where surface grinding was not conducted
between heat treatment and wear testing. Figure 5.5 shows microhardness measurements taken through
the depth of each alloy both before and after surface grinding. Table 5.8 additionally reports retained
austenite content at the surface of each sample before and after surface grinding, measured via XRD.
As observed in Figure 5.5, all non-ground alloys displayed a drop off in hardness values close to
the edge of the sample before leveling off about an average bulk hardness value. This “drop off” is
indicative of decarburization at the surface of the sample as a decrease in C leads to a decrease in the
hardness of martensite. Though the level of decarburization varies in degree between alloys, it likely
played a role in the wear data recorded for the non-surface ground samples. The drop in hardness of this
suspected decarburization layer was about 150 HV for the 0.01 wt pct Nb alloy, 100 HV for the 0.25 and
0.5 wt pct Nb alloys, and only about 25 HV for the 1.0 wt pct Nb alloy, suggesting that the improvement
in wear resistance of the higher-Nb alloys may be associated with the reduced extent of decarburization.
The decarburized layer ranged in depth from about 125 - 350 μm. The depth, however, may have varied
slightly as a function of the position of the first indent with respect to the edge of the sample which is
additionally influenced by the resolution of the light optical microscope on the microhardness indenter.
If it assumed that the 1200 grit surface grind removed 0.1 mm of the decarburization layer for the
samples that were not machined, and wear was not constant over the entire 12.9 cm2 exposed paddle face,
but rather decreased linearly from the paddle end to the sample holder (as was observed), this predicts a
volume of 15 - 130 mm3 (0.1 - 1.0 g) that was affected by decarburization. These estimations also support
the wear data in Figure 4.34(a) in that the sharpest drop in hardness for the 0.01 wt pct Nb alloy shows the
most significant mass loss The difference in mass following Bond abrasion for the machined and not
machined 0.01 wt pct Nb alloy was 0.1 g (11 mm3), indicating that the entire volume of material removed
was within the decarburization layer. This is likely the source of the increased wear observed for this test
as a reduction in C is associated with a decrease in hardness and wear resistance.
102
|
Colorado School of Mines
|
(a) (b)
(c) (d)
Figure 5.5 Vickers microhardness measurements taken near the surface of (a) 0.01, (b) 0.25, (c) 0.5,
and (d) 1.0 wt pct Nb heat treated samples. Measurements were taken both before and after surface
grinding. Dotted line represents the average of the bulk hardness data.
Table 5.8 Retained Austenite Measurements Before and After Surface Grinding
Retained Austenite (vol pct)
Nb (wt pct)
No Surface Grind Surface Ground Difference
0.01 22.1 24.6 + 2.5
0.25 21.0 25.2 + 4.2
0.5 22.5 27.0 + 4.5
1.0 19.4 24.8 + 5.4
Once hardness measurements were repeated after the samples were consistently surface ground,
this hardness drop was no longer observed as per Figure 5.5. The existence of a decarburization layer is
supported by the retained austenite measurements in Table 5.8 All alloys showed an increase in retained
austenite after surface grinding. When considering austenite stability, an increase in C content at the
surface of the surface ground samples corresponds to a decrease in M temperature. Therefore, it is
s
103
|
Colorado School of Mines
|
expected that the sample surface may also contain a decreased martensite volume fraction and increased
austenite volume fraction after surface grinding, as shown in Table 5.8. However, it is still of importance
to note that the retained austenite content measurements made via XRD likely vary by at least ± 5 vol pct
and a definitive trend is difficult to confirm.
The micrographs of the worn paddle end surfaces in Figure 4.38 highlight the degree of surface
deformation that occurred as a result of the Fe ore impacting the steel samples. XRD was attempted to
characterize the retained austenite fraction in this deformed layer by scanning the worn surface directly,
cutting away a thin layer from the surface (approximately 1 mm) and scanning the back side of the worn
face, as well as lightly grinding the worn layer until flat and thinning the sample with HF to remove the
deformation layer imposed by grinding. The rough worn surface produced diffraction patterns that had
very high noise, as well as extremely broad peaks, leading to inconclusive retained austenite
measurements. Example spectra are shown for each alloy in Figure 5.6. It is observed that these
diffraction patterns result in peaks ranging in broadness at their base from 2θ values of 2 ° up to
almost 7 °. This can be compared to the XRD spectrum in Figure 4.7 for the surface and center of the
grinding ball where the peaks were generally only 2 ° - 3 ° in width. It is also noticed that for the α-[200]
(65.0 °), γ-[220] (75.2 °), γ-[311] (91.4 °), and α-[220] (99.0 °) peaks especially, it is difficult to
distinguish between peaks and background noise.
α
γ
α
γ
α γ γ α
Figure 5.6 Diffraction patterns obtained via XRD on worn surfaces following Bond abrasion testing.
Spectra are included for the 0.01 (bottom), 0.25, 0.5, and 1.0 (top) wt pct Nb alloys.
The second method in which the backside of the worn surface was scanned again resulted in
retained austenite measurements outside of the deformed zone as the interaction volume of X-rays only
reached a depth of approximately 10 μm [76], whereas the deformed layer was likely closer to
100 - 1000 μm below the surface that was scanned. The final method consisting of grinding and thinning
104
|
Colorado School of Mines
|
the worn surface was again unsuccessful as the combination of grinding and thinning likely removed the
entire deformed layer imposed by the Bond abrasion test. If any of this layer did remain, it is highly
unlikely it was greater than 10 μm and X-rays were expected to penetrate deeper into the bulk
microstructure. Overall, in order to confidently quantify the retained austenite in the TRIP layer, EBSD or
neutron diffraction would be necessary.
5.7.2 Dry Sand/Rubber Wheel Abrasion Data
DSRW wear testing was employed to evaluate scratching resistance of metallic materials. Data in
Figure 4.43 showed that wear resistance increased as a function of Nb content by approximately 0.02 g
per every 0.1 wt pct Nb alloyed. However, the relationship between Nb content and mass loss was not
linear, but rather the effect of Nb on wear resistance appeared to decrease as Nb content is further
increased. Since the retained austenite volume fraction remained approximately constant among alloys, it
is perhaps likely that resistance to scratching wear was related to the surface hardness of the material, best
shown by Figure 4.44(b) in which mass loss is plotted as a function of sample hardness.
Figure 4.45 indicates an uneven wear track with wear extending longer on the right side
compared to the left side. However, this non-uniformity was both not significant in comparison to
examples provided in ASTM Standard G65 and was consistent across all samples, so was not considered
to be of concern. Micrographs in Figure 4.46 show a very thin deformed layer at the worn surface of the
DSRW samples. EBSD or neutron diffraction would be necessary to better quantify the phase
characteristics of the wear layer.
105
|
Colorado School of Mines
|
CHAPTER 6
SUMMARY
The wear resistance of niobium-alloyed grinding media containing eutectic niobium carbides was
investigated. Four alloys with a constant baseline composition of
Fe-0.98C-0.96Mn-0.44Cr-0.26Cu-0.24Si-0.12Ni-0.024Mo-0.021S-0.014P-0.01Sn-0.005V (wt pct) and
Nb additions of 0.01, 0.25, 0.5, and 1.0 wt pct were laboratory prepared and heat treated in salt pots to
create a microstructure of martensite and retained austenite, similar to industrial grinding balls. Wear
testing was conducted using Bond abrasion and dry sand/rubber wheel (DSRW) testing. The primary
conclusions of the present study are outlined as follows:
• The as-cast laboratory prepared material consisted of a primarily pearlitic matrix with
NbC eutectic carbide networks in the alloys with 0.25 wt pct Nb and greater. As Nb
content was increased, the volume fraction of the eutectic-containing NbC constituent
increased as well.
• The as-cast laboratory prepared ingots were hot rolled at 1085 ºC to simulate the hot
rolling procedure of the industrial bar stock used for the grinding balls. The
microstructure revealed a favorable response to hot rolling with the NbC networks
effectively broken up, providing a more uniform distribution of the Nb carbide. There
still existed NbC rich and NbC deficient bands. Small voids were visible adjacent to large
NbC particles; however, no macroscopic defects or fracture occurred as a result of hot
rolling.
• NbC solubility calculations were performed to aid in the understanding of the
solidification behavior of Nb and C in austenite. It was determined that Nb may remove
up to 0.13 wt pct C from austenite for the 1.0 wt pct Nb alloy in order to form NbC,
leading to an expected decrease in the matrix hardness of 6.4 HV (0.2 HRC). Using
empirical equations from literature, an increase in M temperature of 21.4 °C was
s
predicted to result from depleting the austenite of 0.13 wt pct C. Dilatometry performed
on hot-rolled plates with increasing Nb revealed an increase in hardenability as Nb
content was increased. 1.0 wt pct Nb additionally was observed to increase the M
s
temperature of the steel by 29 °C.
106
|
Colorado School of Mines
|
• Salt pots were used to scale up the heat treatments determined via dilatometry on larger
samples for Bond abrasion and DSRW testing. Samples were austenitized to 875 °C, held
for 1200 s, quenched to 150 °C, and allowed to air cool to room temperature. The
purpose of this heat treatment was to mimic the hardness observed in the industrial
grinding balls in the experimental lab prepared material. With a minimum surface
hardness goal of 61 HRC, the heat treatment design was successful in obtaining a
hardness of 64 HRC for the 0.01 wt pct Nb alloy. Hardness increased further as Nb
content was increased. A second goal of the heat treatment included matching a
microstructure consisting of approximately 30 vol pct retained austenite. However, only
about 20 vol pct retained austenite was measured in the experimental material and no
distinct trend was observed between retained austenite content and Nb alloying.
• Bond abrasion testing revealed similar wear resistance as Nb alloying was increased,
though hardness of the samples tested increased by approximately 0.5 HRC per
0.25 wt pct Nb alloyed. SEM on worn surfaces did not show any obvious differences in
the wear patterns between alloys.
• Dry sand/rubber wheel results show a favorable decrease in wear with increasing Nb. A
decrease was observed with Nb content up to a 65 pct reduction in mass loss between the
0.01 and 1.0 wt pct Nb alloys. It can thus be concluded that Nb content is favorable in
terms of scratching-abrasion resistance. These data also suggest that a different wear
mechanism may be dominating in Bond abrasion versus dry sand/rubber wheel wear
tests.
107
|
Colorado School of Mines
|
CHAPTER 7
FUTURE WORK
Extended work pertaining to this study could initially include additional microstructural
characterization both on the NbC particles as well as on the matrix. In terms of additional matrix
characterization, electron backscatter diffraction (EBSD) could be employed to better quantify relative
amounts of austenite and martensite in heat treated alloys, as well as in the deformed layer following wear
testing. Transmission electron microscopy (TEM) would be beneficial in characterizing if cementite
and/or bainite exist in the microstructure after heat treatment. Further characterization and understanding
of NbC could also be explored. C extraction replicas could aid in understanding relative amounts of Nb in
NbC among the different alloys. Electrochemical dissolution has potential in providing a better evaluation
of NbC fraction.
Further mechanical testing would also be of interest. Nanoindentation could provide an accurate
evaluation of carbide and matrix hardness. Scratch testing would also provide a qualitative assessment on
how different alloys respond to scratching abrasion in terms of the active wear mechanism (gouging,
plowing, cutting, etc.). As ball mills create a considerable degree of impact, Charpy impact testing would
be beneficial to evaluate the toughness of the experimental material and compare toughness values to
current grinding balls. Fracture analysis could then be performed to determine if fracture is likely to
propagate along carbide networks. Tensile and work hardening properties of grinding ball materials could
also be determined.
An additional dilatometry study to evaluate untransformed austenite as a function of quench
temperature could be useful. Current industrial grinding balls contain a critical amount of retained
austenite that aids in both wear resistance as well as toughness properties. Quench temperatures other than
room temperature and 150 °C may be explored first via dilatometry then scaled up in the salt pots, with
retained austenite measurements performed via XRD and/or EBSD. Additional C partitioning treatments
could also be explored to increase austenite stability.
Alloy design may be another area of interest in future studies. Alternative Nb contents both below
0.25 wt pct and above 1.0 wt pct could be evaluated. As Nb content is increased, there is a possibility
NbC transitions from a eutectic to a primary structure. Jarreta and Wright found an increase in wear
resistance in slurry pump impellers due to an addition of 15 wt pct primary Nb carbides to an
A05 steel [44], so an evaluation of grinding media may be beneficial to determine the optimal Nb content
associated with the highest wear resistance. Other alloys of interest may incorporate a reduction in C
levels from that of the base alloy and a fixed Nb content to evaluate the effect of C content on wear
108
|
Colorado School of Mines
|
CHAPTER 2
MAPPING
Borehole tunneling machine usually drills the hole while recording direction and length.
Becauseofthispropertyofboreholes, mineralobservationsalongtheboreholeshaveaformat
as set of lines containing mineral amounts observed. Each line has starting point, length,
azimuth angle and inclination angle. To apply tensor based imputation models, we need
to map this boreholes in continuous spherical coordinate to tensor in discrete Cartesian
coordinate.
To be specific, we should populate a tensor-like data structure, in our case array, with
the number of observations and mineral amounts observed along the boreholes. If multiple
observations should be mapped into a single grid point of array, the amounts of mineral will
be averaged out. After mapping, output array will be 4-dimension(x, y, z, a kind of mineral).
This mapped output array will be input for imputation model.
This mapping affects on quality of imputed output, so we provide four mapping methods,
and each mapping has it’s own pros and cons on accuracy, computation time, and density.
That means good mapping method should result that the distance between mapped position
and borehole position should be small, computation time should be small, and the number of
populated points after mapping should be large. Here are the descriptions for each mapping.
Our data is in 4-dimensional space, but for simplicity, here we plotted example figures in
2-dimensional plane.
3
|
Colorado School of Mines
|
where X ∈ Rn×m is data matrix and M ∈ {0,1}n×m is mask matrix indicating observations.
After imputation, we will stack these 2D matrices up to 4D tensor.
Figure 3.1: Joint Schatten p-norm and l -norm robust matrix completion for missing value
p
recovery
3.2 Low-Rank Tensor Completion with Spatio-Temporal Consistency
Low-RankTensorCompletionwithSpatio-TemporalConsistencyorSTItriestominimize
rank of data at a specific 2D plane and difference between consecutive planes. This methods
works well when data has consistency along the axes like as video data. Mineral distributions
are usually continuous, so this method is applicable. To apply STI on 4-dimensional tensor,
we need to split this into 3-dimensional(x,y,z) tensors for each mineral. And then along the
consistency axis, we will solve below minimization problem for each plane X
i
n n
min kX k +α kX −X k2
Xi|n
1
i ∗ i+1 i F (3.2)
i=1 i=1
X X
s.t.|X (k,h)−D (k,h)| ≤ ǫ,∀(k,h) ∈ Ω ,∀i
i i i
where (k,h) ∈ Ω denotes entries observed, and D (k,h) are values of them. kX k denotes
i i i ∗
trace norm of matrix X , and kX −X k2 denotes Frobenius norm squared of X −X .
i i+1 i F i+1 i
α and ǫ are hyper-parameters, but ǫ is adaptive so that it will be decreased as we have
10
|
Colorado School of Mines
|
3.3 GAIN with p-norm minimization
We combinated with Generative Adversarial Imputation Nets or GAIN[1] with Schatten
p-norm minimization[3]. GAIN is neural network based imputation model, and it two neural
networs, those are generator and discriminator. Generator tries to generate vector good to
foolthediscriminator,anddiscriminatortriestodistinguishwhichentriesaregenerated(fake)
and are came from input(real).
3.3.1 Input and Output
3.3.1.1 generator
We can flat tensors of any dimension into 1-dimensional vector. Let the length of input
vector is n. Given the sparse input value vector X ∈ Rn, mask vector M ∈ Rn, and random
noisevectorZ ∈ Rn, generatorwillpreprocesstheinputbyfillingitoutvaluesofun-observed
entrieswithnoisevectorZ; M⊙X+(1−M)⊙Z ∈ Rn, andconcatenatethisfilledvaluevector
with mask vector M, and it will be final input. And then Generator generates recovered
vector by feeding this input to generator’s neural network. So the generator function will be
g : (M ⊙X +(1−M)⊙Z)×M → G ∈ Rn
3.3.1.2 discriminator
Given the generated vector G ∈ Rn and mask vector M, we will preprocess M into hint
vector H ∈ Rn by removing observed entries of M with probability p so that H will
hint
contains partial information in M. Then we will mix real vector X with generated vector G
to M ⊙X +(1−M)⊙G.
From the concatenation between M ⊙X+(1−M)⊙G and H, discriminator will return
probability vector D ∈ [0,1]n whose each entry represents the probability that this entry is
real. So the discriminator function will be
d : (M ⊙X +(1−M)⊙G)×H → D ∈ [0,1]n
12
|
Colorado School of Mines
|
Note that this is different from loss function of discriminator of original GAIN[1],
n
(M · log(D (G,M)) + (1 − M ) · log(1 − D (G,M)))
i=1 i i i i
−
n
P
where
• We only consider entries which is not included in hint vector H because it is too naive
for discriminator to distinguish whether entry which is determined as observed by H
is real or fake(it is always real). So we excluded observed entries in H in loss function.
• We differently weighted observed entries and unobserved entries. If we weights both
equally and if number of observed entries and unobserved entries are unbalanced, in
other words if input is too sparse or dense, discriminator will fall into mode collapse.
For example, suppose input is very sparse; density of observed entries are less than
1%. Then there is naive but powerful strategy for discriminator that discriminator
just predicts all entries are fake. Then this naive ’all-fake’ prediction will be correct
for about 99% entries, and it will be wrong for only about 1% entries, thus loss will
be very low. As discriminator falls into mode that this predicts all entries are fake,
generates cannot learn anything from discriminator, and generator suffers from mode
collapse too, generating same outputs G regardless of input X.
So we weights more for sparse one and less for dense one. For example, if input is very
sparse, then in equation (3.4), the term n (M −H ) will be very small, thus term
i=1 i i
n ((M − H ) · log(D (G,M))) will bePweighted more. On the other hands, term
i=1 i i i
Pn (1−M ) will be very large, thus term n ((1−M )·log(1−D (G,M))) will be
i=1 i i=1 i i
Pweighted less. P
3.3.3 Neural Network setting
Generator and Discriminator both uses fully-connected dense layer, so if input size n
is large, then computation cost will be expensive. For example, suppose our geological 4-
14
|
Colorado School of Mines
|
dimensional(x, y, z, a kind of mineral) input is tensor of size 80 × 80 × 80 × 3, then the
number of neurons in first layer will be 80 × 80 × 80 × 3 = 1,536,000 which is unrealistic.
And cardinality of training set will be 1 which is obviously too small to train model.
Thus we split-ed 4D tensor into small 4D cubes such as cube whose size is 10 × 10 × 10
× 3 = 3,000 with strides (2, 2, 2) so that we have reasonable number of neurons 3,000 in
first layer and enough cardinally of training set 1,000. So generator will output small cube
given another small cube, and discriminator will output small cube with probabilities given
another small cube also. So the number of neurons in last layer should be exactly half of
number of neurons in first layer described in subsubsection 3.3.1. If size of split-ed cube is
large enough, then we can still exploit information of spatial structure of input.
The number of neurons except for first and last layer, number of layers, and activation
function for each layer are hyper-parameter. But as activation function, we used rectified
linear unit except for last layer and sigmoid for last layer, for generator and discriminator
both. Thus generated output will be in range [0,1]n but this will cover range Rn because we
applied min-max scaler in preprecessing step.
Figure 3.3: modified GAIN network
15
|
Colorado School of Mines
|
3.3.4 Training
Trainingisupdatingoftheweightsandbiasesmatrices. Bothgeneratoranddiscriminator
have their own neural network each. We may use simple gradient descent method to update
weights and biases of neural networks following,
∂loss
Wg,n ← Wg,n −η g
ij ij g ∂Wg,n
ij
∂loss
Wd,n ← Wd,n −η d
ij ij d ∂Wd,n
ij
(3.5)
∂loss
Bg,n ← Bg,n −η g
j j g ∂Bg,n
j
∂loss
Bd,n ← Bd,n −η d
j j d ∂Bd,n
j
where Wg,n, Wd,n are weight of i’th row and j’th column of n’th layer’s weights matrix
ij ij
and Bg,n, Bd,n are j’th bias of n’th layer, and η , η are learning rate for generator and
j j g d
discriminator each. But in practice, Adam optimizer which is modified version of above
gradient descent methods works better, so we used Adam. We used TensorFlow to execute
tensor operations such as singular value decomposition or multiplication or subtraction on
tensors.
3.3.5 Experiment on MNIST dataset
To evaluate the result numerically, we calculated normalized mean absolute error or
NMAE as in [[4], [5]]
|M −X |
(i,j)∈Γ ij ij
NMAE = (3.6)
|Γ|(r −r )
P max min
We tested on MNIST dataset following same way as paper [1]. Below are resulted examples.
Inthisfigure,fifthrowissparseinputwhoseeachobservedentryisconservedwithprobability
called sampling rate, first row is mixed input M ⊙X +(1−M)⊙Z ∈ Rn with noise vector
Z described in paragraph 3.3.1.1. Second, third, and fourth row are generated samples. We
wrote NMAE of sparse input → NMAE of recovered output.
16
|
Colorado School of Mines
|
3.4.1 Recovery of mask tensor
We can distinguish which entries are observed by using mask tensor, but we cannot
distinguish after imputation which entries can be regarded as observed other than observed
entries in mask tensor. So on top of recovery in data tensor, we need to recover mask tensor
also.
3.4.1.1 Low-Rank Tensor Completion with Spatio-Temporal Consistency
Input of Low-Rank Tensor Completion with Spatio-Temporal Consistency above is
f : X ×M → X ∈ Rn×m×l
recovered
where X ∈ Rn×m×l is data tensor and M ∈ Rn×m×l is mask tensor. Instead of giving data
tensor and mask tensor, we can provide mask tensor and mask tensor so that we can recover
mask tensor not the data tensor.
f : M ×M → M ∈ Rn×m×l
recovered
After impuation of mask tensor, we will set the entries of mask tensor as observed(value is
larger or equal than 1) whose values are above threshold t which is hyper-parameter, so
s
that
0, if M ≤ t
i s
M i = 1, if t s < M i ≤ 1
M , otherwise
i
for i = 1,...,n.
23
|
Colorado School of Mines
|
3.4.1.2 GAIN with p-norm minimization
We can exploit the discriminator of GAIN to recover mask tensor. Actually output of
discriminator is itself prediction for mask tensor. So if discriminator is trained well, we can
get reasonably good mask tensor. We can also set hyper-parameter threshold for GAIN, t ,
g
so that we can earn mask tensor following
M , if 1 ≤ M
i i
0, else if D(G,H) ≤ t
M = i g
i
1, else if t < D(G,H) ≤ 1
g i
D(G,H) i, otherwise
for i = 1,...,n. In addition to get recovered mask tensor, we can filter output of generator
following below
X , if 1 ≤ M
i i
X i = 0, else if D(G,H) i ≤ t g
G(X,M) , otherwise
i
for i = 1,...,n. This filtering c an be used in simple GAIN described above subsection 3.3
too.
3.4.2 repeating GAIN with p-norm and Low-Rank Tensor Completion
In our experiment, after each repetition, density is increased by about 40%. So we can
repeat both alternatively until we get result with satisfactory density d .
s
24
|
Colorado School of Mines
|
Algorithm 1 Alternative imputation
1: procedure repeatedGAINSTI(data tensor X, mask tensor M)
2: d ← density of M
c
3: X , M ← X, M
gain gain
4: while d ≤ d do
c s
5: X , M ← low rank tensor completion with X , M
sti sti gain gain
6: X , M ← GAIN with p-norm minimization on X , M
gain gain sti sti
7: d ← density of M
c gain
return X , M
gain gain
3.4.3 Experiment on geological data
WeexperimentedalternatingGAINwithp-normminimizationandLow-Rankcompletion
method on gold(Au) of area of our data. Our data is mapped into 80×80×80×1 tensor with
inexact mapping. In this experiment, we did not sample the observed entries to calculate
the error between imputed values and observed values, and just tried to recover unobserved
entries. Because we have no ground truth values on unobserved entries, we cannot calculate
NMAE, but this experiment is to see whether our model is able to generate plausible figure.
In this experiment, we alternated GAIN and STI 3 times. The output array of one of both
will be input array of one of another. The split-ed cube size of GAIN is 6×6×6 and we gave
stride as (1, 1, 1).
We provides 3-dimensional results and 2-dimensional results. In 2-dimensional results,
we provides 2-dimensional slices randomly selected from 3-dimensional tensors, and X mark
indicates entries not recovered, △ mark indicates entries recovered, and ◦ indicates entries in
input data. The color intensities indicate gold amounts. The left image is before imputation,
and the right image is after imputation.
25
|
Colorado School of Mines
|
3.4.3.1 Interpretation on results of GAIN and STI alternation model
Before imputation, the data density which is the proportion of entries regarded as ob-
served by mask tensor, is increased from 0.89% to 6.29%. But as you can see in above ,
our model can recover entries in the area already dense in input data, but can not recover
entries in the area sparse. So we need to improve our model to capture global structure to
impute entries in the area sparse. And when we tested our model on data with sampling
rate is 70%, we found that this model failed to decrease NMAE.
3.5 Convolutional Generative Adversarial Imputation Networks
As you can see in above, GAIN and STI alternation model is not able to impute sparse
spacefarfromdensespace. ThisisbecauseSTIisgoodatimputingentriesbetweenobserved
entries, but not good at imputing entries in the space where there is no spatio-temporal con-
sistency. And GAIN cannot capture global structure because input cube size is limited. So
we developed CGAIN(Convolutional Generative Adversarial Imputation Networks) to uti-
lize information of the larger space than GAIN. The idea is that we attach the convolutional
boxesaroundthespacewewanttoimpute(SOI).LikewiseinGAINwewillvectorizetheSOI,
and in addition we will append the convolutional results to SOI, as illustrated in Figure 3.15.
By applying convolutions, we can utilize information of larger space while increasing
the length of input vector relatively smaller than that of GAIN. Because the far the space
from SOI, the less important the space, we can increase the size of convolutional kernel of
the boxes far from SOI. On top of that, we can place the convolutional boxes for different
directions to capture the trends of mineral distributions for each direction, as illustrated in
Figure 3.16.
28
|
Colorado School of Mines
|
3.6 Convolutional Normalized Mean Squared Error
To evaluate the results better, we introduces new metric Convolutional Normalized Mean
SquaredError(CNMAE).Whenthepositionsofimputedvaluesareslightlyslipped, CNMAE
canevaluatebetterthanNAMEasillustratedFigure3.17. CalculatingCNMAEwillbesame
as calculating NMAE, but the only difference is that the error is calculated between window-
wise values instead of point-wise values. So the error for each window will be the difference
between the average of imputed values and the average of observed values only for observed
positions within a window.
Input Input
Mask Tensor Amount Tensor 𝑁𝑀𝐴𝐸
1 0 0 1 3 0 0 9
7 + 9+ 1+ 1
=
0 1 0 0 0 7 0 0 Imputed
6 ∗ (9− 1)
1 0 1 0 1 0 4 0 Amount Tensor
= 0.375
0 1 0 0 0 1 0 0 3 0 9 0
Sampled Sampled 7 0 0 0
Mast Tensor Data Tensor 0 1 4 0 𝐶𝑁𝑀𝐴𝐸
0 0 1 0
1 0 0 0 3 0 0 0 0 + 0+ 1+ 1
=
0 0 0 0 0 0 0 0 6 ∗ (10− 1)
0 0 1 0 0 0 4 0
= 0.037
0 0 0 0 0 0 0 0
Figure 3.17: When the positions of imputed values are slightly slipped, window-wise error
stays small as it should be
30
|
Colorado School of Mines
|
3.7 Experimental results of CGAIN
As we did in subsection 3.4, we alternated CGAIN and STI 3 times to get the denser
results. Here we sampled the input tensor with two different methods. For a given sampling
rate p ,
s
• Sample observed entries with probability p , that is regular sampling.
s
• Sample bulks of space whose volume is proportion p of total volume, that is irregular
s
sampling.
Insecondsamplingmethod, observedentrieswillbespreadless. Wesetthesamplingrate
as 70% likewise GAIN experiment. After sampling, the observation density becomes 0.8%.
About the error, we earned 3 improved results from 4 cases. For regular sampling, NMAE
is increased from 15.764 to 19.800 and CNMAE is decreased from 0.00555 to 0.000525. For
irregular sampling, NMAE is decreased from 8.631 to 8.593 and CNMAE is decreased from
0.000304 to 0.000284. The density of imputed entries is increased from 0.8% to about 50%.
The input is 4D tensor with 3D gold distribution and 3D silver distribution, so if a
relation exists between gold and silver distribution, our neural networks may capture it. In
this experiment, the size of space of interest is 6×6×6. We added 4 convolutional boxes
for +x, -x +y, and -y directions with width 3 and kernel size is (3, 3, 3) and added 2
convolutional boxes for+z and -z directions with width 2 and kernel size is (2, 2, 2). Here
we also provide the result as 2D slices and 3D space.
31
|
Colorado School of Mines
|
ABSTRACT
Due to the demand for rare earth elements for everyday technology and applications, there has
been much research initiated into the extraction and recovery of rare earth elements. An otherwise
unknown mineral, eudialyte, is a zirconium silicate consisting of rare earth oxides, specifically the heavy
rare earth oxide yttrium (III), with only trace amounts of thorium and uranium. The focus of this research
project was to investigate and develop a beneficiation and leaching procedure for processing the Norra
Kärr eudialyte ore. The development of the type of beneficiation and leaching experiments conducted
was aided by a review of different physical separation methods and the treatment of iron and silica in
other industries.
After mineral characterization, a two-stage beneficiation process was developed, consisting of
gravity and magnetic separation. The gravity separation portion comprised of preliminary heavy liquid
separation tests done using both sodium polytungstate and methylene iodide at different size fractions.
Different size fractions were studied for liberation purposes. This gravity separation step was
implemented for the removal of the heavy iron-bearing mineral aegirine. This float product is then
processed in a wet high-intensity magnetic separation (WHIMS) at 1 Tesla to separate the paramagnetic
eudialyte from the non-magnetic gangue minerals. The implementation of this process resulted in limited
success for a clear separation of eudialyte from its gangue. The overall results yielded no significant
upgrade of eudialyte using the beneficiation process proposed. However, the proposed process did show
that iron could be rejected through either gravity or magnetic separation, a definite benefit for further
hydrometallurgical treatment.
After the conclusion of the beneficiation tests, hydrometallurgical testing was done. The samples
used in these leaching experiments were non-magnetic concentrates, where most of the iron was rejected
via WHIMS. Two separate leaching processes were investigated to eliminate or minimize the formation
of silica gel within the solution, while still recovering the total rare earth elements (TREEs). The first
leaching process treated the concentrate in a 0.1 M solution of sulfuric acid at 25, 50 and 75°C at two and
four-hour intervals. This leaching process resulted in gelation of the leach liquor as well as filtrate
solution, but recovered the TREEs and Zr. The second leaching process limited the amount of water and
acid available to the concentrate by only adding enough concentrated sulfuric acid to completely wet the
sample. The acid-wet samples were then left for 30 minutes, one hour (then oven dried) or air dried
before leached with DI water. While no gelation was observed during or after this leaching process, little
to no rare earth elements and zirconium were recovered. It has become evident through these
beneficiation and leaching experiments, that a generalized method, applicable in many other mineral
processing industries for commonly known minerals, may not be the best method for processing
iii
|
Colorado School of Mines
|
INTRODUCTION
The aim of this research project was to investigate an advanced beneficiation and
hydrometallurgical process applicable to the Norra Kӓrr eudialyte ore with the final purpose being to
extract the yttrium and other rare earth elements. This research was supported by the Critical Materials
Institute (CMI) which is a multi-institutional, multi-disciplinary energy innovation hub of the U.S.
Department of Energy. The goal of CMI is to target certain technologies that make efficient use of its
materials and eliminate the need for materials subjected to supply disruptions, through collaborative
innovations between industrial partners, national laboratories and academic institutions. CMI defines five
critical elements and two near-critical elements essential for the competitive clean energy industry in the
United States. The five critical elements CMI focuses on are: terbium, europium, dysprosium,
neodymium and yttrium; as well as, the two near-critical elements: lithium and tellurium. These critical
and near-critical elements are so defined because they a) provide essential and specialized properties to
advanced products or systems, b) have no easy substitutes and c) are subject to supply risk. [1] CMI’s
approach to the critical materials problem can be summarized in four groups of research:
▪ Diversifying supplies: relying on more than just one source.
▪ Developing substitute materials that can meet needs without using the materials we use
today.
▪ Using the available materials more efficiently to reduce waste in manufacturing processes
and increase the adoption of recycling.
▪ Forecasting which materials might become critical in the future. [1]
The research conducted in this project is concentrated in the diversifying supply group and
advanced beneficiation subgroup. The main goal of the advanced beneficiation subgroup is to develop
new sources of critical materials by establishing an efficient beneficiation process applicable to critical
element-bearing minerals. Approaching this goal requires exploring minerals not previously research
comprehensively, with considerable critical element source potential. Eudialyte is one such mineral due
to its relatively unknown status, but high reserve quantities. Eudialyte is a potential source for yttrium
and other rare earth elements with the added advantage of low concentrations of the radioactive elements
thorium and uranium.
The first step in many extractive processes is to try to create an enriched preconcentrate that has
been removed of unwanted materials or materials that are valuable for another process. This beneficiation
step is not only important for the future hydrometallurgical processing of the ore, but economically as
1
|
Colorado School of Mines
|
LITERATURE REVIEW
This chapter provides background information pertaining to previous and current processing
techniques for the beneficiation and hydrometallurgical treatment of rare earth bearing minerals. The
information obtained by conducting this literature review was essential in the development and execution
of the experimental design set forth.
2.1 The Rare Earth Elements
The term “rare earth elements” refers to the 17 metallic elements comprising of the lanthanides,
yttrium and scandium. [2] These elements have been referred to as a group because of their chemically
similar properties. This group can be further divided into the yttrium heavy and cerium light rare earth
elements subgroups based on the chemical similarity within the group. The light rare earth group consists
of the first eight elements of the lanthanide series (atomic numbers 57 – 64) and sometimes scandium.
The heavy rare earth group consists of the rest of the elements in the lanthanide series (atomic numbers 65
– 71) and yttrium.
2.2 Rare Earth Element Applications
In modern technology, the rare earth elements are in demand and considered of great importance.
Major application areas include magnets, catalysts, electronics, glass, ceramics and metal alloys. The
proportion of world total rare earth consumption in each category is summarized in the graph below.
Proportion of Total Rare-Earth Consumption in 2010
Other
Ceramics
6%
6%
Glass
Electronics
24%
7%
Metal Alloys
18%
Magnets
20%
Catalysts
19%
Figure 2.1. Proportion of total rare earth consumption in 2010. [2]
3
|
Colorado School of Mines
|
Yttrium and cerium are the rare earth elements of importance with respect to eudialyte, since they are the
most abundant in the mineral. Yttrium is essential for fluorescent light phosphors, computer and
television displays, automotive fuel consumption sensors and microwave filters, as well as to stabilize
zirconia in thermal plasma sprays used on the surfaces of aerospace components to protect them from
high temperatures. [1,3] Cerium can be used in a variety of applications, including: as a polishing agent in
precision optical polishing of glass, mirrors, optical glass and disk drives; as a sensitizer in ceramics; in
catalytic converter, and many other areas. Although the applications of only two rare earth elements were
named, the importance of the availability of all rare earth elements should not be understated. With the
consumption of rare earths expected to continue to grow, especially in the energy, electronics and
optoelectronics sectors, demand for these elements is also expected to rise in accordance. [4] These
elements and their compounds are necessary for the development of many modern technological devices
that consumers have become heavily dependent on in a daily basis.
Unlike their given group name, these elements are considered abundant in the earth’s crust, 240
ppm in total rare earth abundance in comparison to the abundance of carbon at 200 ppm.
RE: Lanthanides + Y + Sc
CM: Cu + Ni + Pb + Zn + Sn
1000
)
m
p
p
(
ts
100
u
r
c
s 'h 10
tr
a
e
e
h 1
t
n
i
e
c
n 0.1
a
d
n
u
b
A 0.01
RE CM Ni Zn Ce Cu Nd La Y Sc Pb Sn Tm Cd Hg Ag
Figure 2.2. Abundance of elements in the earth’s crust. [2]
2.3 Rare Earth Element Bearing Minerals
Although considered abundant in the earth’s crust, these elements are not found in their elemental state in
nature. [2,5] They can be found in many rock formations, usually in the form of oxides, silicates,
carbonates and phosphates. [6] Rare earths can be found in over 200 minerals, however, about 95% of all
rare earth resources occur in just three minerals, in consecutive order starting with the mineral most rich
in rare earths: bastnӓsite, monazite and xenotime. This does not include rare earths found in ion-
4
|
Colorado School of Mines
|
adsorption clays. [65] While these minerals are considered prime candidates as resources for rare earth
elements, eudialyte also has the potential for becoming such candidate due to its higher heavy rare earth
element concentration in comparison to these conventional rare earth minerals. Eudialyte also exhibits
very low concentration of radioactive elements as another benefit as a rare earth resource. [7]
2.3.1 Bastnäsite
Bastnäsite is a fluorocarbonate mineral with a rare earth concentration of about 70% rare earth oxide
(REO), primarily consisting of cerium and little to no thorium, thus considered to be a primary source for
the light rare earths. Two major deposits for bastnäsite are Bayan Obo, China and Mountain Pass,
California, USA. The chemical composition is The density varies between 4.90 –
5.20 g/cm3 [9] and is paramagnetic. [1,5] Gravity and magnetic separation techniques have been used to
(cid:4666)(cid:1844)(cid:1831)(cid:1831),(cid:1829)(cid:1857)(cid:4667)(cid:4666)(cid:1829)(cid:1841)(cid:2871)(cid:4667)(cid:1832).
beneficiate the mineral, with flotation considered to be the most relied upon using a fatty-acid or
hydroxamate-based collector system. [8]
2.3.2 Monazite
Monazite is a phosphate mineral that contains approximately the same amount of REO content as
bastnäsite at 70%, however, unlike bastnäsite, monazite has a higher concentration of the radioactive
elements thorium and uranium. REO content is primarily made up of cerium, lanthanum, praseodymium
and neodymium. The chemical composition is . The density varies between 4.98 – 5.43
g/cm3. [1,5,9] Well known monazite deposits are in Van Rhynsdorp and Naboomspruit in South Africa, in
[(cid:4666)(cid:1844)(cid:1831)(cid:1841),(cid:1846)ℎ(cid:4667)(cid:1842)(cid:1841)(cid:2872)]
Bayan Obo in China and in Colorado, USA.
2.3.3 Xenotime
Xenotime is a yttrium-bearing mineral containing about 67% REO, mostly consisting of just the
heavy rare earth elements. In many instances, it is found alongside with monazite and beneficiation
techniques focus on separation from monazite through flotation and magnetic separation. [5,10] The
chemical composition is and the density varies between 4.40-5.10. Xenotime deposits can be found
in placer cassiterite deposits in Malaysia, Indonesia and Thailand, as well as the heavy mineral sand of
(cid:1851)(cid:1842)(cid:1841)(cid:2872)
Australia.
2.3.4 Ion-adsorbed clays
Primarily found in southern China, ion-adsorption clays have been mined since the 1970s and are the
world’s most important resource for the heavy rare earth elements. These clays have developed in
morphologically predisposed areas, by lateritic weathering of felsic rocks deposits that contain rare earth
element bearing minerals. The ion-exchange phenomena present in these clays consists predominately of
cation exchange on the layer surfaces of the clays and chemisorption of anions at the edges of the layer.
5
|
Colorado School of Mines
|
During the adsorption process, heavy rare earth elemental cations are preferably adsorbed onto the clays
due to their higher charge and size. [11] Ion-adsorption clays require little to no prior beneficiation
processing before hydrometallurgical treatment, making them excellent candidates, both industrially and
economically, as a source for heavy rare earth elements.
2.3.5 Eudialyte
Eudialyte is a zirconium silicate mineral, notable for its high concentration of the heavy rare earth
elements, specifically yttrium. The crystal structure comprises of a nine-membered silica ring and a six-
membered ring of calcium octahedra that is held together by zirconium octahedra and three-membered
silica rings. The general chemical composition that characterizes eudialyte is as follows:
. The density of the mineral varies between 2.70-3.10 g/cm3.
[9] Typical gangue minerals associated with eudialyte are aegirine, nepheline syenite and feldspar. Table
(cid:1840)(cid:1853)(cid:2872)(cid:4666)(cid:1829)(cid:1853),(cid:1829)(cid:1857)(cid:4667)(cid:2870)(cid:4666)(cid:1832)(cid:1857),(cid:1839) (cid:4667)(cid:1852)(cid:1870)(cid:1845) (cid:2876)(cid:1841)(cid:2870)(cid:2870)(cid:4666)(cid:1841)(cid:1834),(cid:1829)(cid:1864)(cid:4667)(cid:2870)
3.1 displays the chemical composition for the specific eudialyte sample used for this project as analyzed
through MLA (Mineral Liberation Analysis). Specific gravity and magnetic properties are also given in
Table 3.1. Both eudialyte and aegirine are considered paramagnetic, however, the magnetic susceptibility
of aegirine is treated as being greater than that of eudialyte’s because it is a predominately iron-bearing
mineral. However, due to the zeolite crystal structure of the eudialyte minerals, there is a variety of
different compositions that can still be identified as a eudialyte group mineral. This mineral usually
forms in alkaline igneous rocks, such as the nepheline syenite of the Ilimaussaq complex in the southwest
of Greenland. The zirconsilicate mineral has also been found at Pajarito in New Mexico, USA. Other
deposits can be found in former regions of the USSR and Canada: such as the Khibina and Lovozero
complexes in Russia and the Mont Saint-Hilaire complex in Canada. [5,13,14] The eudialyte mineral is of
special interest due to some of the advantages it has over traditional sources of rare earth elements. These
advantages include its very low concentrations of thorium and uranium, as well as its ability to be readily
dissolved in acid. The name eudialyte is derived from the Greek phase meaning “well decomposable.”
[9]
The eudialyte mineral used in this project originates from the Norra Kärr deposit in southern Sweden.
The Norra Kärr deposit is a zirconium and rare earth element enriched peralkaline nepheline syenite
intrusion which hosts the eudialyte group minerals. The deposit has been found to contain three
compositional varieties of the eudialyte mineral. These three groups are as follows: 1) iron rich, REE
poor from lakarpite, 2) iron and manganese bisected, heavy REE rich from pegmatitic grennaite and 3)
manganese rich, light REE rich from migmatitic grennaite. [15]
6
|
Colorado School of Mines
|
2.4 Physical Beneficiation Techniques
This section details common techniques used in the beneficiation of rare earth minerals. These
techniques include gravity and magnetic separation. While it was not specifically employed in this
project, a discussion on froth flotation is included because it is a frequently used method in many mineral
processing plants all over the world. Previous physical beneficiation techniques conducted on eudialyte
are also examined in this section as they pertain to the experimental design of this project.
2.4.1 Gravity Separation
Beneficiation of rare earth minerals, or any minerals for that matter, can be done by exploiting the
different specific gravities of the minerals present within the ore mined. If the mineral of interest has a
specific gravity vastly different than the specific gravities of the gangue minerals present, as well as being
sufficiently liberate, the choice of beneficiation technique is easy to make with gravity separation. Over
the years different gravity separation instruments have been developed, such as jigs, sluices, spirals,
shaking tables, fine particle separators and cyclones. Before any gravity work should be done, it is
important to know if the specific gravity differential is sufficient. This can be done by calculating the
concentration criterion. This simple mathematical equation does not consider differences in particle sizes
and assumes good liberation of all minerals within the sample. The equation is as follows:
Equation 2.1 [16]
(cid:4666)(cid:1830)ℎ−(cid:1830)(cid:3033)(cid:4667)
(cid:1829)(cid:1829) =
(cid:4666)(cid:1830)(cid:3039) −(cid:1830)(cid:3033)(cid:4667)
Where CC is the concentration criterion, D is the specific gravity of the heavy, light or fluid components
as denoted by the subscript h, l and f, respectively. When the absolute value of CC is greater than 2.5,
there is potential for some form of gravity concentration down to 200 mesh. If the absolute value of CC
is between 2.5 – 1.75, separation is effective to 100 mesh. A CC value between 1.75 – 1.50, separation is
possible to 10 mesh with some difficulty. A CC value between 1.50 – 1.25 can yield separation to ¼
inches, also with some difficulty. Finally, if the absolute value of CC is less than 1.25, the potential for
gravity concentration is virtually impossible with the use of commercial techniques, however, a
separation can still be achieved through heavy media/liquid separation. [16,17] Using water as the fluid
medium with an SG of 1.0, D (aegirine) of 3.55 and D (eudialyte) of 2.9, the CC would be 1.34.
h l
Another preliminarily gravity separation method to determine the potential an ore has for effective
separation is sink/float analyses. Sink/float analyses or heavy media/liquid separation is done by putting
the sample in a liquid whose density is between the two densities one wishes to separate. Separations are
made to develop the standard washability curves used to estimate the reaction of a sample to gravity
concentration. A partition curve can also be constructed to evaluate the effectiveness of a specific
concentration method or instrument. [17] Traditionally, hazardous organic liquids were used to achieve
7
|
Colorado School of Mines
|
densities far greater than that of water’s. Three of the most commonly used organic heavy liquids are
bromoform, tetrabromoethane (TBE) and methylene iodide, producing densities of 2.9, 3.0 and 3.2 g/cm3
respectively. [18] However, due to toxicity of these chemicals, friendlier substitutions have been research.
One such substitution is the aqueous solution sodium polytungstate. A density of 3.1 g/cm3 can be
achieved at 20°C by dissolving the sodium polytungstate powder in water until saturation is reached.
Recovery of the sodium polytungstate can be done through evaporation. [19]
A discussion on gravity concentration would not be complete without limited details regarding some
of the older and newer instruments that have been used. Jigs are one of the oldest methods used to
concentrate coarse material that is close in size or if the differential in densities is large, a wider size range
may also yield a good concentration. [20] The particles are presented to the jig bed consisting of a screen
that is fluidized. The pulsating water results in a suspension of particles. Once the pulsating ceases, the
particles settle according to specific gravity, allowing the heavier particles to sink and form a concentrate
underflow, while the lighter and smaller particles form a tailing overflow. [21] Another well-known
concentration method are spirals. As the material travels through the spiral, gravitational and centrifugal
forces act on the particles, separating coarse light particles from fine heavy ones. Additionally, shaking
tables have been widely used throughout the mining industry during the cleaning stages since they have a
low capacity. Capacity can be increased if multiple-deck tables are used. The separation is driven by
how the differences in specific gravity and sizes respond to an inclined rippled table that oscillated back
and forth. The result is a concentration of fine heavy particles to be collected at the uppermost section of
the table, while coarse light particles will be collected at the bottom edge of the table. An illustration of a
shaking table can be seen below.
Figure 2.3. Shaking table schematic. CONS: fine, heavy particles; MIDS: intermediate particles;
TAILS: coarse, light particles. [21]
8
|
Colorado School of Mines
|
Finally, a description of fine particle separation instruments is given. Such separators primarily
utilize centrifugal forces in generating a good concentration since the feed usually involves fine to very
fine particles where the effects of particle size dominate a gravitational separation. The Knelson
concentrator is an inclined bowl lined with perforated ridges to allow for fluidization of the material.
Once the centrifugal force is applied the lighter material is collected through an overflow, while the heavy
material will concentrate at the ridges of the bowl. The Falcon concentrator is another spinning fluidized
bed. The heavier particles migrate to have contact with the bowl walls, while the lighter particles are
collected in the overflow. Although similar in principle, differences between the Knelson and Falcon lie
in design parameters. For example, in the Knelson, the material is directly introduced into the fluidization
zone; while in the Falcon, the material enters a segregation zone along the cone wall where the heavier
particles travel through a bed of gangue to reach the wall of the bowl. This bed of materials is composed
of a lower layer of the heavier particles and an upper layer of gangue, and thus become the segregated
material that will enter the fluidization zone. [22]
2.4.2 Magnetic Separation
Another common technique for the beneficiation of rare earth minerals can be done through
magnetic separation, where the magnetic susceptibilities of the minerals are used. Materials are
considered either magnetically ordered or not, according to the orbital and spin motion of electrons,
which may or may not result in a magnetic moment within the material. Ferromagnetic, ferrimagnetic
and antiferromagnetic materials have a positive magnetic susceptibility and retain permanent
magnetization without the presence of an external magnetic field. Paramagnetic materials may also
exhibit a positive susceptibility due to the presence of unpaired electrons in partially filled orbitals.
However, a magnetic moment is only induced when an external magnetic field is applied but it will not
hold that magnetic moment if the field is removed. Diamagnetism is a basic component of all matter and
a material is classified as diamagnetic when this force cannot be overcome by any attractive magnetic
moments. When an external magnetic field is applied to diamagnetic materials, a repulsive force is
induced, opposing the applied field due to the negative susceptibility. [23,24]
Many rare earth minerals are paramagnetic due to the electron configuration of the rare earth
elements present in the mineral. The rare earth elements have electrons occupying a shielded 4f sub-shell
and the existence of unfilled 4f shells will produce these magnetic properties. [25,26] Magnetic recovery
is dependent on the magnetic field gradient, the applied magnetic field strength, the magnetic
susceptibility of the mineral and the fluid medium. This can be represented by the following equation:
9
|
Colorado School of Mines
|
Equation 2.2 [27]
(cid:1856)(cid:1828)
(cid:1832) = ( (cid:3043)− (cid:3040))(cid:1834)
(cid:1856)
Where F is the magnetic force, V is the volume of the particle in (m3), χ and χ is the volume magnetic
x p m
susceptibility of the particle and the fluid medium, respectively, H is the magnetic field strength in (A/m)
and dB/dx is the magnetic field gradient in (N/Am2). [27,28] It is also important to note how the size of
the mineral particles play a role in the effectiveness of the separation. Gravitational, magnetic and fluid
drag forces each have different dominating effects based on the size of the particle. Fluid drag forces are
proportional to the radius, r, magnetic forces are proportional to r2 and gravitational forces are scaled to
r3. From this relationship, it can be concluded that as the particle size increases, the force due to gravity
become more prominent than on smaller particles, where fluid drag forces dominate. [27]
Magnetic separators can be categorized into four functional groups: dry- low and high intensity
and wet- low and high intensity. Low intensity separators will typically operate at magnetic field
strengths of 0.2 Tesla (2000 gauss) or less, effectively collecting ferromagnetic materials. High intensity
separators can be operated above 0.5 Tesla (5000 gauss) and can efficiently obtain paramagnetic
materials. [29] The following diagram is taken from Norrgran and Mankosa depicting a decision tree for
separator selection.
Figure 2.4. Magnetic Separation Decision Tree. [29]
Low intensity dry-drum magnetic separators are very effective at producing a clean non-magnetic product
or concentrating a magnetic product. The separator consists of a stationary, shaft-mounted magnetic
10
|
Colorado School of Mines
|
circuit enclosed by a rotating drum. Magnetic material is attracted to the drum shell and will unload when
it is rotated out of the magnetic field. The non-magnetic material, on the other hand, will discharge in a
natural trajectory over the rotating drum. A schematic of the dry drum is shown below.
Figure 2.5. Schematic of Dry Drum Magnetic Separator. [29]
As seen in figure 2.4, three different types of dry high intensity magnetic separators can be used. Rare
earth drum utilizes rare earth permanent magnetics to provide a higher magnetic field strength. The
design incorporates a center magnetic element pole that consists of a series of axial poles of alternating
polarity. Steel interpoles are placed between each magnetic pole which concentrate the magnetic flux,
producing a high magnetic gradient at the surface of the drum. [29] Similar to the rare earth drum, the rare
earth roll employs a high magnetic field strength to effectively remove weakly magnetic materials. The
rare earth roll is composed of neodymium-iron-born permanent magnet disks that are wedged between
steel poles. A schematic of the rare earth roll is below.
11
|
Colorado School of Mines
|
Figure 2.6. Schematic of the Rare Earth Roll Magnetic Separator. [29]
The last dry high intensity magnetic separator discussed is the induced roll. The induced roll generates its
magnetism with an electromagnet is almost solely used for mineral sands and industrial mineral
applications. The roll is composed of alternating ferromagnetic steel and non-magnetic rings. The
material is introduced onto the roll and travels through a gap between the roll and the electromagnetic
pole, where the non-magnetic particles will be discharged through a normal trajectory. Paramagnetic or
other weakly magnetic material attach to the roll and are deflected to another collection location.
Similar to the dry drum case, a low intensity wet drum and high intensity rare earth wet drum is
employed. The wet rare earth drum also allows for collection and recovery of weakly magnetic materials
contained in a slurry. The wet low intensity drum is used in many heavy media and iron ore applications.
The design consists of a rotating drum in a tank where the magnetic portion of the drum is in contact with.
When the slurry is introduced into the tank, magnetic materials attach to the drum via magnetic attraction,
while the non-magnetic material is collected in an underflow as is displayed below. The final two
magnetic separators discussed are used in applications where the material consists of fine particles. The
difference between a wet high intensity magnetic separator (WHIMS) and a wet high gradient magnetic
separator (HGMS) is how the direction of the slurry flow is aligned. In a WHIMS, the direction of the
slurry flow is perpendicular to the line of magnetic flux, while in a HGMS, the flow direction is parallel.
A laboratory WHIMS diagram is included below.
12
|
Colorado School of Mines
|
Various rare earth ores contain a mixture of non-magnetic and magnetic minerals, whether those
be ferromagnetic or paramagnetic depends on the composition. In many cases, a combination of different
magnetic separation techniques is used. One such case is the physical beneficiation of coarse heavy
mineral sands from Congolone, Mozambique, where magnetite is removed through a low intensity
magnetic separator due to its ferrimagnetism.1 After the magnetite is removed, the non-ferrimagnetic
material proceeds to a WHIMS unit. To recover rare earth minerals, high intensity magnetic separators
are logically used because of their ability to retrieve paramagnetic material. [2] An example includes the
selective separation of paramagnetic monazite its non-magnetic heavy gangue minerals zircon and rutile.
[30,31]
2.4.3 Electrostatic Separation
Electrostatic separation utilizes differences amongst the conductivities of the minerals present within
the ore. Modes of recovery of similar to those of magnetic separators, in which the particles are subjected
to an electric field (static and/or ionic) and those that become electrically charged are then separated from
those that did not charge. This type of particle charging is done through induction in an electric field.
Conducting particles will polarize such that negative charges will orient towards the positive electrode,
while positive charges will align towards the negative electrode. A material can be classified as a
conductor, non-conductor or semiconductor due to its electrical resistivity and dielectric constant.
Conductors have small resistivity values of about 10-5 ohm·cm and extremely large dielectric constants.
On the other hand, non-conductors have large resistivity values on the order of 1014 ohm·cm. Finally,
semiconductors are materials with properties that lie between those of conductors and non-conductors.
Usually having a resistivity value between 1 and 104 ohm·cm. [32] The following figure shows how a
conducting particle will pass through a drum separator. Conducting particles have a low electron affinity
and will give up electrons to the hopper through contact, resulting in a particle with a positive charge.
Since the rotating drum is positively grounded, the positive particles are now repulsed by the drum and
fall with a gravitational force trajectory. Non-conducting particles have a large electron affinity and will
become negatively charged via the hopper contact. This leads to an attractive force between the
positively grounded drum and the negatively charged particles. The attraction allows the particles to stay
fixed onto the drum.
1 Ferrimagnetism is not to be confused with ferromagnetism in terms of the physics of magnetite; however, for the
purposes of mineral processing, ferrimagnetic materials are processed as ferromagnets due to their comparable
magnetic susceptibilities.
14
|
Colorado School of Mines
|
Figure 2.9. Conducting particles travelling through electrostatic separator. [32]
This mechanism can be seen in the figure below. Typically, electrostatic separation is not widely used and
is only applicable where other beneficiation techniques cannot be used. Electrostatic separators found in
the mineral processing industry generally fall under a drum type or free fall design. Drum separators
operate either a conductance field, ionic field or a combination of the two. Many plants that process
heavy mineral sands find this technique valuable for the separation of rutile2 from monazite and zircon.
[33] The beneficiation of rare earth minerals monazite and xenotime may also apply an electrostatic
separation, since in many cases, the gangue minerals associated with these minerals, such as ilmenite,
have similar specific gravities and magnetic properties. [31] Unlike most other beneficiation techniques,
electrostatic separation requires the processing feed to be completely dry. This condition may lead to
excess energy costs since drying is an expensive unit operation, especially when applied on an industrial
scale.
2 The electrostatic response of rutile is prominent at higher temperatures, greater than 200°C. [33]
15
|
Colorado School of Mines
|
Figure 2.10. Non-conducting particles travelling through electrostatic separator. [32]
2.4.4 Froth Flotation
Froth flotation is one of the most widely used beneficiation techniques in mineral processing due to its
versatility in design parameters that allows for selectively. The basic principle of flotation is the
separation of hydrophobic materials from hydrophilic ones. The mineral of interest to be separated is
made hydrophobic through the addition of surfactants or collectors. These chemicals are
thermodynamically selective to adsorb to the surface of the mineral particles. The mineral particles are
then able to bind to air bubbles and float to the surface of the slurry to be collected. The process requires
a slurry suspension, a selective collector (if the mineral surface is not hydrophobic) and a frothing agent
to promote the formation of bubbles. The theory behind a successful flotation process lies in the
thermodynamics of the mineral surfaces, adsorption and wetting. A low energy state is desired
throughout the process between the mineral particle surfaces and the bubble-particle contact. A
simplified three-phase system is shown below and the condition for a low energy state can be met through
the following equations.
16
|
Colorado School of Mines
|
Figure 2.11. Schematic representation of the equilibrium contact between an air bubble on a solid
immersed in a liquid. [34]
Equation 2.3 [34]
= + cos
Equation 2.3 refers to Young’s equation for a three-phase contact between a solid, gas and liquid, where
γ , γ , and γ are the surface tension energies for solid-gas, solid-liquid and liquid-gas interfaces,
SG SL LG
respectively, and θ is the contact angle formed between the three-phase junction. A large contact angle
results in greater hydrophobicity and a greater potential for flotation. [34] There is an associated change
in free energy when the solid-liquid interface is replaced by a solid-gas interface given by Dupre’s
equation:
Equation 2.4 [34]
Δ(cid:1833) = −(cid:4666) + (cid:4667)
Where ΔG is the change in Gibbs’ free energy. Combining Young’s and Dupre’s thermodynamic
equations yields an expression for the free energy change:
Equation 2.5 [34]
Δ(cid:1833) = (cid:4666)cos −(cid:883)(cid:4667)
Equation 2.5 shows that there is a free energy decrease through the attachment of a mineral particle
surface to an air bubble. It is worth noting, however, that both Young and Dupre’s equation carry
assumptions in their development. Dupre’s equation does not take into account other energy consuming
effects, while Young’s equation is valid in an ideal system at equilibrium with no gravitational effects.
[34,35]
As mentioned before, the advantage of flotation is its ability to change design parameters for
selectively and the choice for surfactants allows for such selectively. The adsorption mechanism
17
|
Colorado School of Mines
|
employed during flotation is important for the success of an effective recovery of the mineral of interest.
Two different classes of adsorption exist, physical and chemical adsorption, and are determined by the
surface forces present. [36] Knowledge of the electrical potential at the surface of the particle and the
electrical double layer is needed for calculation of the adsorption caused by the electrostatic forces.
Fuerstenau and Somasundaran state that a solution containing charged particles must also be electrically
neutral so as to contain an equal amount of oppositely charged ions. However, these oppositely charged
ions are not uniformly distributed, instead located near the surface, creating a Stern plane, as seen in the
figure below. The potential at the Stern plane determines the maximum adsorption, but it cannot be
measured experimentally. Instead, a potential measurement is taken at the shear plane, called the zeta
potential. As a particle moves through an electric field, the liquid nearest the surface of the particle
moves at the same velocity as the particle, while the liquid farther from the surface remains static. It is
the distance between the moving and static liquid that describes the shear plane. [37]
Figure 2.12. Schematic of double electrical layer. [37]
A determination of the type of adsorption mechanism present can be done using several
measurements, such as the zeta potential and adsorption isotherms. Physical adsorption may be taking
place if a cationic collector adsorbs onto the particle surface in a region where the zeta potential is
negative or vice versa. On the other hand, if a cationic collector adsorbs onto the surface while the zeta
potential is positive and negative when an anionic collector adsorbs, chemical adsorption may be taking
place. Physical adsorption is due to weak van der Waals forces, resulting in a low heat of adsorption,
18
|
Colorado School of Mines
|
nonselective reversible behavior and multilayer coverage. In contrast, chemical adsorption is stronger
due to valence forces, resulting in a larger heat of adsorption, selective irreversible adsorption behavior
and monolayer coverage. [38,39]
Adsorption isotherms (constant temperature) show the amount of adsorbate on the absorbent as a
function of the pressure or concentration. According to IUPAC, there are six types of isotherms.
Figure 2.13. Schematic of the six types of adsorption isotherms. [40]
In Figure 2.13, Type I isotherms are characteristic isotherms for microporous materials with no multilayer
adsorption. Types II and III isotherms represent multilayer adsorption in non-porous solids. Types IV
and V show capillary condensation in mesoporous solids, while Type VI shows stepped adsorption.
It is vital to study these variety of parameters before implementing a flotation process on an
industrial scale. Microflotation is such a precursor to understand the response of different reagents on
either pure minerals or ores. Bench scale flotation is the successor to microflotation, usually conducted in
a laboratory setting and is considered predictive of how the flotation process will perform on an industrial
level.
2.4.5 Previous Physical Beneficiation Techniques Conducted on Eudialyte Minerals
This section previews previous beneficiation techniques and results conducted on other eudialyte
minerals, including the Norra Kärr eudialyte mineral. A literature search regarding the physical
beneficiation of eudialyte yielded limited gravity, magnetic and flotation work. Ferron and Rawling
summarize the laboratory work done on the Ilimaussaq eudialyte mineral in “Recovery of Eudialyte from
a Greenland Ore by Magnetic Separation.” Eudialyte samples were taken from three different locations in
19
|
Colorado School of Mines
|
the deposit with different rock types: marginal pegmatite, kakortokite and lujavrite. Furthermore, samples
were taken from three layers within the kakortokite rock type. A combination of dry and wet magnetic
separation was conducted and two flowsheets were developed. The first flowsheet looked at processing
the ore through low intensity magnetic separation, followed by high intensity. The second flowsheet
started with a high intensity unit, followed by the low intensity separator. A low intensity magnetic
separator was used to eliminate the arfvedsonite/aegirine iron bearing minerals that would predominating
respond to the low intensity field. The high intensity magnetic separator was used to reject the non-
magnetic nepheline syenite and feldspar minerals, while collecting the eudialyte bearing concentrate in
the magnetic fraction. The recovery of rare earth oxides was not tracked in these experiments, instead,
zirconium oxide was used as an indicator of the recovery of eudialyte. The results showed that to produce
an acceptable separation, the ore needed to be ground finer than 28 mesh. Although recoveries were
shown to be in the 80s, significant upgrade in the zirconium oxide and the eudialyte could not be achieved
past an upgrade ratio of 2. A heavy liquid separation test was also done using methylene iodide and
acetone mixtures. The goal being to separate the eudialyte from the nepheline syenite/feldspar and
arfvedsonite /aegirine, with specific gravities of 2.8-3.0, less than 2.8 and greater than 3.2, respectively.
A spiral gravity concentration test followed, yielding no significant selectively for the concentration of
eudialyte. [41]
There is little known about the flotation characteristics of eudialyte, since previous experiments
conducted on eudialyte are limited. Russian literature reports eudialyte recovery via flotation using
sodium oleates and oleic acid as collectors. [42,43] Ferron, Bulatovic and Salter conclude that the use of
amphoteric collectors depends on the eudialyte composition, pH and conditioning time. The following
flowsheet was developed.
Figure 2.14. Double reverse gangue flotation for processing a REO-eudialyte ore. [43]
20
|
Colorado School of Mines
|
Flotation experiments were also done on the eudialyte ore from the Lovozero deposit at the Kola
Peninsula in the former USSR. On average, the eudialyte ore contains 13.5 wt % ZrO and 2.5 wt% rare
2
earths. Like the flowsheet in Figure 2.14, a reversible flotation flowsheet was developed, where fatty acid
collectors were used to first float aegirine. The eudialyte containing tails would then go to a eudialyte
flotation where monoalkylphosphates were used as collectors. [44]
Magnetic and flotation beneficiation work was also conducted on the Norra Kärr eudialyte
mineral by RWTH Aachen University. Dry and wet high intensity magnetic separators were used with
the goal to produce a separation between the major gangue and eudialyte in one step. However, the
magnetic susceptibilities between aegirine and eudialyte overlap enough to hinder the ability to create a
clean separation. Focus shifted to flotation concentration, with the goal being, again, to avoid a two-step
flotation process, as the one suggested for the Lovozero eudialyte mineral. Three different eudialyte
samples were tested as raw ore feeding into the circuit, while three other eudialyte samples were pre-
concentrates from a magnetic separation step. Overall, the pre-concentrate eudialyte samples used in the
flotation step yielded the highest upgrade ratios and recoveries in the 80s. [45] Stark, Silin and Wortuba
conclude that a selective direct flotation for eudialyte can be achieved using a mixture of
mono/diphosphoric acid esters as collectors, and oxalic acid and sodium hexametaphosphate as
depressants, at a pH below 4.
The Norra Kärr project in Sweden was undertaken by Tasman Metals Ltd., in consultation with
ANZAPLAN, with the intention on determining the most suitable beneficiation route for the Norra Kärr
mineralized material. Different techniques were investigated, such as spiral concentration, electrostatic
separation, high-G separation, magnetic separation and froth flotation. [46] Results showed that aegirine
could be selectively floated, but co-flotation of non-liberated particles concluded that a direct flotation of
eudialyte would be unsuccessful. [47] High recovery values were recorded for eudialyte via WHIMS, but
with no significant upgrade in the rare earth concentration. [48,49]
The literature survey regarding eudialyte beneficiation experiments indicate that at least a
multiple step process is necessary for separation of the eudialyte mineral from its gangue components.
2.5 Hydrometallurgy of Rare Earth Element Bearing Minerals
Hydrometallurgy is a chemical processing technique involving the use of aqueous chemistry to
extract metal from an ore or other materials. The three major areas associated with hydrometallurgy are
leaching, concentration and purification, and metal recovery. This section will discuss leaching and
separation processes.
21
|
Colorado School of Mines
|
2.5.1 Leaching
Leaching is the process by which metals are converted to soluble salts in an aqueous solution. Types
of leaching reactions include: water solvation, acid or alkali dissolution, base exchange, complex ion
formation, oxidation and reduction of mineral. Depending on the concentration of rare earth
concentrations, typical leaching processes used on major rare earth minerals, such as monazite or
xenotime, consist of a caustic soda or sulfuric acid leach. [50]
2.5.2 Leaching of Silicate Materials in Industry
Silicate materials, such as eudialyte, exist all over the world and in many forms, affecting many
different aspects of industry. This section will discuss how silica is treated or removed in major
operations, such as geothermal wells and zinc silicate minerals for the production of zinc. Typically,
silica removal technology involves aging the solution at a certain temperature, resulting in complete silica
polymerization and colloidal particles. A coagulant, such as lime, is added and the resulting flakes are
then separated in a settler. [67]
Geothermal wells produce steam or hot water (that may be flashed at a lower pressure to produce
steam) serve as sources for electric and thermal energy. These wells can be found in Mexico, New
Zealand, Indonesia, El Salvador, Japan, California, New Mexico, Nevada and Idaho. [66] Many of the
waters obtained from these brines are saturated with silica, that once in solution at a high saturation, has
the potential to form colloid and gelatinous mass. The issue of silica in the waters is specific to the
operational features of obtaining the waters. The features are as follows: the brine flows into a producing
well and loses pressure and temperature, causing it to partially evaporate. The vapor is then separated and
directed to a turbogenerator. The liquid phase is taken through heat exchangers for extraction of heat and
then reinjected into the geothermal reservoir to prolong the time of the geothermal field with compliance
of environmental regulations. [67] It is when the solution is reinjected into the wells while decreasing in
temperature, that the solution becomes overly saturated with silicic acid. Once the ortho-silicic acid is
formed, the polymerization to a gelatinous mass is almost instantaneous and its rate can be described by
equation 2.6.
Equation 2.6 [68]
(cid:1856)(cid:1829) (cid:3041)
− = (cid:1863)(cid:4666)(cid:1829)−(cid:1829)(cid:3032)(cid:4667)
(cid:1856)(cid:1872)
Where C is the concentration of monomeric silica at time t, and Ce is the equilibrium solubility of
amorphous silica at a temperature. The polymerization rate is proportional to the concentration of
hydrogen ion below pH 1, will proceed more rapidly at elevated temperatures and in turbulent
environments. [68] This gel is highly viscous and deposits on the equipment, hindering their efficiency, as
well as the energy process efficiency from these brines. Silica electrocoagulation is a derivative from the
22
|
Colorado School of Mines
|
To conclude, the issue of solubilized silica is not a relatively new issue and in fact has been
addressed in many different industries. From electrocoagulation in geothermal brines to investigating
various pH ranges in solution, as well as limiting water content, the problem with silica gel has been dealt
with for the appropriate industry. Still, as the issue arises, different methods are being explored to better
understand the silica solution chemistry. Some recommend introducing flocculants or aqueous chromium
in the +6-oxidation state to hinder the polymerization of silica. However, each case is specific and the
solution to the solubilized silica needs to be specific as well.
2.5.3 Leaching of Eudialyte Mineral
As mentioned before, the name eudialyte derives from the Greek word for “well-decomposable”
in acid. However, the issues with the leaching of eudialyte lie with the co-dissolved silica. This silica
forms a gelatinous phase hindering the filtering processing for rare earth element extraction. [58] The
current goal of processing eudialyte is to achieve a reasonable recovery of leached rare earth elements
while minimizing or eliminating the formation of the silica gel. Previous hydrometallurgical tests done
by Lebedev (2003), Lebedev, et al. (2003), and Zakharov et al. (2011), involved the high temperature
leaching with concentrated sulfuric acid followed by dilution of the pulp with a sodium sulfate solution.
This process produced an insoluble residue containing the rare earth element double sulfate salts. The
salts would then be washed with water and recovered by converting the sulfates to nitrates or chlorides.
[58]
There is a discussion regarding the efficiency of leaching the rare earth elements and zirconium
as sulfate or chloride ions in terms of the solubility. Also, which acid minimizes the silica gel formation
when used in a concentrated manner. It has been suggested by Voßenkaul et al., that the recovery of rare
earth elements is more favorable in chloride systems. In using hydrochloric acid, rare earth chloride salts
are developed and are typically more soluble in water than the sulfate salts from employing the sulfuric
acid. The solubility of rare earth element sulfate salts in water decreases proportional to the decrease in
atomic number of the rare earth element, except for cerium and praseodymium. Thus, the heavy rare
earth elements stay in solution, while the light rare earth elements are precipitated. [2] Since yttrium and
the heavy rare earth elements are soluble, double-sulfate precipitation is not possible. Double-sulfate
precipitation is used for separating rare earth elements by their light or heavy respective groups. Equation
2.9 shows the double-sulfate precipitation chemical reaction:
Equation 2.9 [2]
(cid:2871)+ (cid:2870)− +
(cid:884)(cid:1844)(cid:1831)(cid:1831) +4(cid:1845)(cid:1841)(cid:2872) +(cid:884)(cid:1840)(cid:1853) ↔ (cid:1844)(cid:1831)(cid:1831)(cid:2870)(cid:4666)(cid:1845)(cid:1841)(cid:2872)(cid:4667)(cid:2871)∙(cid:1840)(cid:1853)(cid:2870)(cid:1845)(cid:1841)(cid:2872)∙(cid:1834)(cid:2870)(cid:1841)
However, in terms of minimizing the formation of the silica gel, the use of sulfuric acid may have a
greater advantage than the hydrochloric acid. Apart from its low cost, volatility and corrosive activity,
24
|
Colorado School of Mines
|
sulfuric acid has better solubility in water at room temperatures than hydrochloric acid. Concentrated
sulfuric acid can be found at 98 w/w%, while hydrochloric acid strength is 37 w/w%. To minimize the
silica gel formation, the solution must not have access to large amounts of water. The reasoning lies in
the thermodynamics and kinetics of the silica-water system. With the addition of acid to a silicate, shown
by the chemical equation 2.10, the silicate will acidify to form the weak monosilicic acid:
Figure 2.15. Molecule of monosilicic acid.
Equation 2.10 [61]
(cid:4666)(cid:1845) (cid:1841)(cid:2870)(cid:4667) (cid:4666)(cid:1871)(cid:4667)+ (cid:884)(cid:1834)(cid:2870)(cid:1841)(cid:4666)(cid:1864)(cid:4667)↔ (cid:4666)(cid:1845) (cid:1841)(cid:2870)(cid:4667) −(cid:2869)(cid:4666)(cid:1871)(cid:4667)+(cid:1845) (cid:4666)(cid:1841)(cid:1834)(cid:4667)(cid:2872)(cid:4666)(cid:1853)(cid:1869)(cid:4667)
Once the silicic acid is formed, a polymerization reaction occurs analogous to a condensation
polymerization reaction. The presence of water aids in the polymerization process. The polymerization
process is shown below. [59]
Figure 2.16. Polymerization mechanism for the development of silica gel. [59]
The polymerization proceeds forward to maximize the formation of siloxane linkages (Si-O-Si),
essentially forming a gel with internal siloxane linkages and external SiOH groups. [62] To minimize or
eliminate the silica gel formation, it is concluded that the system needs a to be deprived of water during
the acidic leach since the exposure to water is driving the polymerization following the acidification of
the silicate. A recent approach in seeking to prevent the formation of the silica gel involves a “dry
digestion” of the eudialyte mineral with hydrochloric acid. The process provides just enough acid to wet
the mineral sample allowing the silica to precipitate. The amount of acid to “wet” the mineral should be
around the stoichiometric or slightly below that amount. However, due to the small volume available, the
precipitates should grow to larger particles that can be separated from the valuable metals. [61] The
varying parameters in these experiments include varying acid concentration, retention time in acid and
amount of water used to leach the elements. It was concluded that acid concentrations above 3 M HCl
25
|
Colorado School of Mines
|
and retention times over 10 minutes should yield in significant rare earth element recovery without the
risk of gelling during or after the dry digestion.
Another potential risk in the leaching process is the presence of iron in solution. Until the 20th
century, dissolved iron hindered recovery of many common metals, such as zinc, lead and copper. These
metals and other non-ferrous metals would be produced pyrometallurgically, through high temperature
smelting processes. The iron would report to the slag with other consequential impurities.
Hydrometallurgically speaking, these metals could not be produced at comparable recoveries as their
pyrometallurgical counterparts. However, there was still a need to find a method to treat these ores given
the advantages of hydrometallurgy over pyrometallurgy. It was not until the middle of the 20th century
that zinc would be produced electrolytically under the Roast-Leach-Electrowin (RLE) process. The
Jarosite, Goethite and Conversion processes followed soon after, effectively eliminating any obstacles
iron presented in solution. [63] While the metals driving this hydrometallurgical innovation are not
desired in this project’s goals, the lesson of iron dissolution is the same. Eudialyte and some of its gangue
contain significant amounts of iron and when dissolved under an intense acidic environment, will result in
large quantities of iron in solution. Dissolved iron in solution makes future processing of rare earth
element separation difficult as it is difficult to separate the rare earth elements from iron. [64] Therefore,
special precautions should be employed to limit the amount of iron going into the leaching solution so
recovery losses of the rare earth elements are minimized.
2.5.4 Separation Processes for Rare Earth Oxides from Solution
The following will briefly discuss common separation techniques for separating the individual rare
earth elements from solution of rare earths.
Selective Oxidation
Cerium, praseodymium and terbium are the rare earth elements that can be separated through
selective oxidation due to their occurring trivalent and tetravalent oxidation states. The natural occurring
state of cerium is Ce(III) and it can be removed from the rare-earth mixture by oxidizing to Ce(IV). The
removal of Pr(IV) and Tb(IV) is brought about via precipitation in an aqueous solution since their
tetravalent states are not stable in the aqueous solution. [2]
Selective Reduction
Trivalent samarium, europium and ytterbium elements can be separated through reduction to their
divalent state. Marsh et al. used a buffered acetate solution to separate these rare-earth elements by
reductive extraction into a dilute sodium amalgam. It is known that Sm, Eu and Yb metals cannot be
obtained through the metallothermic reduction of their halides. Therefore, during a mixture of rare earth
halides and calcium, the Sm, Eu and Yb are not reduced, but instead remain in the slag where they can
later be separated. [2]
26
|
Colorado School of Mines
|
Fractional Crystallization
Fractional crystallization exploits differences in solubilities to bring about a crystallization of the
least soluble component through evaporation. Double ammonium nitrates have been used for the
separation of lanthanum, praseodymium and neodymium, while double magnesium nitrates are used for
samarium, europium, gadolinium, as well as the ceric group. To separate yttric group elements, bromates
and ethyl sulfates can be applied. Other applied chemicals include a sodium rare earth EDTA salt for
separating gadolinium, terbium, dysprosium and yttrium; and a rare earth hexa-antipyrine iodide salt for
the separation of erbium, thulium, lutetium and yttrium. [2,14]
Fractional Precipitation
Fractional precipitation is the removal of one or more of the rare earths from solution by the
addition of a chemical to form a less soluble compound, and differs from fractional crystallization since
no other compound is added to the solution. Double sulfates and hydroxides are commonly used in
addition to double chromates. [14]
Ion Exchange
The method of ion exchange involves the exchange of ions between an electrolyte solution and an
ion exchanger or resin. An aqueous solution containing the metal is passed through a bed of solid organic
resin in particulate form. Through an adsorption stage the metal ions load or adsorb onto the exchanger.
Following adsorption, an elution stage allows the ions to desorb from the exchanger. An ion of higher
charge will displace one of lower charge or if the charges are similar, the ion with the larger radius will
replace the smaller radius ion. [1] The most useful complexing agents applied at EDTA and HEDTA
(hydroxyethyl-ethylene-diamine-triacetic acid). Apart from separating Eu-Gd, Dy-Y and Yb-Lu pair,
EDTA is effective at separating most rare earths from each other.
Solvent Extraction
Solvent extraction is the selective transfer of ionic species from an aqueous solution to an immiscible
solvent, such as an organic solution. The aqueous and organic solutions come into contact with each
other, where the metal ions and the organic form a compound that is more soluble in the organic phase,
effectively transferring the metal ions to the organic phase. The extraction of the pure metal from the
organic phase involves the introduction of another aqueous phase, splitting the metal/organic compound.
In the rare earth industry, the use of the solvent extraction is an extremely favored method since relatively
simple equipment is needed to achieve a highly pure metal. [1,14]
27
|
Colorado School of Mines
|
EXPERIMENTAL DESIGN AND PROCEDURE
This chapter details the experimental design developed following characterization and mineralogy
data of the Norra Kärr eudialyte mineral used for these experiments. Also, descriptions of the
beneficiation and hydrometallurgical experiments are provided.
3.1 Norra Kärr Eudialyte Mineral Characterization
As mentioned before in section 2.3.5, Table 3.1 shows the specific gravity and magnetic
properties of non-specific eudialyte minerals and associated gangue, as well as the chemical composition
of the Norra Kärr eudialyte and gangue minerals.
Table 3.1. Eudialyte and gangue mineral characteristics. [9,12]
Mineral Specific Gravity Chemical Composition Magnetic Properties
Eudialyte 2.70-3.10 Na (Ca,Ce) (Fe,Y,Mn)ZrSiO (OH,Cl) Paramagnetic
4 2 22 2
Aegirine 3.50-3.60 NaFeSi O Paramagnetic
2 6
Potassium Feldspar 2.50-2.60 KAlSi O Non-magnetic
3 8
Nepheline Syenite 2.55-2.60 (Na,K)AlSiO Non-magnetic
4
The material obtained for the MLA analysis was a representative sample of the entire eudialyte sample.
This representative sample was acquired by processing the entire eudialyte sample through a Jones riffle
and a rotating turntable splitter. As Holmes states, there is great responsibility resting on a very small
sample, so it is essential that samples are truly representative of the bulk. [51]
Figure 3.1 and Table 3.2 show the particle size analysis results of the eudialyte sample where the P is
80
111.0 microns. Table 3.3 shows that the Norra Kärr eudialyte sample is made up of predominately
silicate minerals at about 65 wt% and the eudialyte mineral is within the 12.1 wt% zirconium minerals.
The MLA image (figure 3.2) shows the eudialyte mineral in bright red and fairly well liberated in this
captured section of the sample. Analysis of the liberation of eudialyte can be found in Figure 3.3. Greater
liberation is shown by the curves approaching the upper right corner of the plot. The best liberation of the
eudialyte mineral in the sample was in the 200x400 mesh fraction. Denoted by the light blue curve, 78%
of the particles in the 200x400 mesh fraction contained 95% or more eudialyte minerals.
28
|
Colorado School of Mines
|
In addition to MLA information, a TOF-SIMS analysis was conducted for supplementary
elemental information.
Figure 3.4 False color TOF-SIMS image of Norra Kärr eudialyte sample: a) Display of Zr/ZrO, Al
and Fe/FeO ions, b) display of Zr/ZrO, Fe/FeO and combined Y/Ce ions. [74]
Since the TOF-SIMS provides elemental information about the sample surface, it was useful in showing
elemental-based mineral associations. In Figure 3.4a, Zr/ZrO, Al and Fe/FeO ions were displayed on the
sample surface in red, green and blue, respectively. The mineral particles displaying those colors
contained those specific ions. Eudialyte is a zirconium-based mineral and any mineral particles shown in
red were considered to be eudialyte. Figure 3.4b shows Zr/ZrO, Fe/FeO and combined Y/Ce ions in red,
green and blue, respectively. This image was valuable in showing the elemental associations between
Zr/ZrO and Y/Ce ions. Again, since eudialyte is a zirconium bearing mineral with rare earth elements, we
expect those elemental ions to “light up” in the same locations. This is most evident in the color overlay
image in Figure 3.4b, where the purple is considered eudialyte with zirconium and the rare earth elements
yttrium and cerium. The TOF-SIMS analysis was used to concluded that the eudialyte mineral was
associated with zirconium, yttrium and cerium.
Before conducting any beneficiation work, a screen analysis test was done to assess if there was
any preferential mineral deportment to certain size fractions. This result indicates a HLST should be done
at different size fractions. The screen analysis test was done by wet sieving the eudialyte sample into the
size fractions shown in Table 3.4. The analysis and percent distribution show the weight percent of
specific elements in the size fractions. Percent distribution values do not vary much with respect to any
of the elemental groups shown. Thus, it is concluded that there is no preferential mineral deportment by
screening the material.
31
|
Colorado School of Mines
|
Table 3.4. Screen Analysis Test Results
Analysis (%) % Distribution
Weight (g) Weight (%) TREE Zr Fe Na, K, Al TREE Zr Fe Na, K, Al
Feed
60.0 0.011 0.045 0.105 0.37
(analyzed)
Feed
45.9 100 0.011 0.042 0.102 0.362 100 100 100 100
(calculated)
100x140 9.7 21 0.010 0.037 0.092 0.379 17.7 18.7 19.1 22
140x200 9.3 20.4 0.010 0.038 0.103 0.364 18.6 18.3 20.7 20.5
200x270 8.7 18.9 0.011 0.041 0.109 0.356 18.9 18.7 20.3 18.6
270x400 10.2 22.3 0.011 0.039 0.105 0.361 22.1 20.7 23 22.2
(-400) 8 17.4 0.015 0.057 0.1 0.346 22.6 23.5 17 16.7
3.2 Description of Beneficiation Experiments
Taking into account the literature survey conducted, as well as the characterization and
mineralogy of the Norra Kärr eudialyte sample, the beneficiation consisted of a two-stage process via
gravity and magnetic separation. The inclination toward this combination of separation techniques
derived from a limited amount of previous work on a process where both gravity and magnetic separation
were used to beneficiation eudialyte. As mentioned in the literature review, before implementing a large-
scale gravity separation instrument, such as the shaking table or Falcon concentrator, a heavy liquid
separation test (HLST) would gauge the potential for and predict larger scale instrumentation. A
preliminary flowsheet of the beneficiation process is below. The goal of the gravity separation/HLST was
to disassociate the major iron-bearing mineral aegirine from the eudialyte and remaining gangue in the
sink and float, respectively.
Figure 3.5. Preliminary beneficiation flowsheet for Norra Kärr eudialyte sample.
32
|
Colorado School of Mines
|
Initially, the non-toxic sodium polytungstate was used for HLST at three different specific
gravities (2.7, 2.95, 3.08) and at the mesh size fractions listed in Table 3.4. After conducting the HLST
with the sodium polytungstate, it was found that a higher specific gravity value would need to be obtained
to effectively separate the aegirine from the eudialyte and other lighter gangue. Another HLST was
administered at Hazen Research, Inc., at a specific gravity of 3.2 with methylene iodide. All the HLST
conducted utilized a centrifuge due to expedite the settling process. The float material was then processed
through the WHIMS for separation between the paramagnetic eudialyte and non-magnetic gangue.
After initial test work was done, an advanced flowsheet developed for investigating different
parameters through beneficiation. HLST were done at four different size fractions3: as-received ore
sample, pulverized sample and screen material at +/- 400 mesh. The sample was pulverized to achieve a
better degree of liberation in comparison to the as-received sample. The ore sample was also screened to
assess a difference between the screened and pulverized material, since pulverizing the sample may have
caused differences in the surface morphology of the sample. The flowsheets are shown below. Finally, a
WHIMS test was done on the as-received sample, the pulverized sample and the +/- 400 mesh samples.
This was done to serve as a baseline WHIMS test and for determining differences within these samples
that had prior processing.
Figure 3.6. Flowsheet for as-received Norra Kärr eudialyte sample.
3 All HLST conducted from this point forward were done at an SG of 3.2 using methylene iodide for best chance of
gravity separation.
33
|
Colorado School of Mines
|
Figure 3.11. Viscosity of aqueous sodium polytungstate solution as a function of density at 25°C. [52]
Once the samples were dry and the three different heavy liquid solutions prepared, approximately four
grams of each size fraction were placed in a 15-mL vial with 10-mL of solution. Five size fractions and
three densities yielded fifteen vials for this experiment. The vials were then placed in a centrifuge to
expedite the settling process for 5 minutes at 4000 rpm. The float and sink material were carefully
removed from the vial and washed with DI water before being allowed to dry. The dried masses for the
individual float and sink portions were recorded. Finally, a small fraction of each portion was taken for
XRF analysis.
Methylene Iodide Heavy Liquid Separation Test
The HLST done using methylene iodide as the media were conducted by Hazen Research, Inc. The
samples sent to Hazen were the as-received ore, the pulverized ore and +/- 400 mesh ore. Before the
samples were sent to Hazen, they were analyzed with Microtrac’s particle size analyzer. Approximately
twenty-five grams of as-received ore, pulverized ore and +/- 400 mesh ore were sent to for the HLST.
Each sample was placed in a centrifuge for seven minutes at 800 rpm as referred to in Figure 3.12. The
dried masses of the float and sink were recorded and sent back for analysis via XRF.
Wet High Intensity Magnetic Separation
A laboratory scale WHIMS (Figure 2.7) was used to process the float material obtained from the
HLST using methylene iodide. A slurry was created by adding the float material to one liter of water. A
screen matrix was used through which the material will pass for magnetic collection. The laboratory
WHIMS shown in Figure 2.7 can be broken down into two pieces, corresponding to the two valves the
material will pass through. The first valve connects the bowl and the chamber that holds the screen
matrix.
36
|
Colorado School of Mines
|
Figure 3.12. Heavy Liquid Separation Flowsheet. [53]
The second valve is the discharge valve after the material as passed through the screen matrix. By
adjusting these valves, one is able to adjust the speed of the material flow through the WHIMS. For these
experiments, the valves were kept half open so there was some retention time of the material in the screen
matrix chamber. The WHIMS was magnetized at 1 Tesla and the slurry was slowly added to the bowl,
where the non-magnetic fraction gradually discharged into a labeled bucket. After all the material cleared
the bowl, the non-magnetic bucket was removed. The magnetism was removed and the magnetic
materials were collected in a magnetic-labeled bucket. Three passes were done to ensure an efficient
separation. Both magnetic and non-magnetic fractions were pressure filtered and dried before XRF
analysis.
3.4 Description of Hydrometallurgical Treatment
The hydrometallurgical treatment of the Norra Kärr eudialyte concentrate consisted of a leaching
process. As mentioned in the literature review, although eudialyte is easily decomposable (as its name
implies), once the silicate mineral is acidified and decomposed, there is a high chance of formation of
colloidal and gelled silica. The goal of these leaching experiments is to extract the rare earth elements, as
well as the zirconium, while minimizing the formation of the silica gel or any colloidal silica particles that
would make the filtering of the leach solution difficult. Two different leaching processes were
investigated. In the first process, leaching of the eudialyte sample was done in an excess of acid available
37
|
Colorado School of Mines
|
to the mineral. Six experiments were done using this process by varying the temperature and leaching
time. In the second process, the eudialyte sample was processed under a starving condition where only
enough concentrated acid to wet the sample was added and later leached with DI water. For this method,
three experiments were conducted by varying the retention time in acid and drying conditions.
A flowsheet of the process is given in Figure 3.12. In each leaching process, a non-magnetic
concentrate eudialyte sample was used. This non-magnetic concentrate sample was produced at a 0.36
Tesla in the WHIMS to separate the aegirine into a magnetic fraction, while leaving the eudialyte and
other gangue in the non-magnetic fraction. As mentioned in the literature review, high concentrations of
iron in the leaching solution is detrimental to the recovery of the valuable materials. By creating a non-
magnetic concentrate, the iron content in the sample is reduced. Sulfuric acid was used in leaching
process 1 to investigate the recovery of rare earth sulfates. Concentrate sulfuric acid was used in leaching
process 2 to minimize water exposure and minimize silica gel formation.
Figure 3.13. Flowsheet for leaching processes for Norra Kärr eudialyte sample.
3.5 Hydrometallurgical Treatment Procedure
This section describes the procedures for leaching process 1 using dilute sulfuric acid, and
leaching process 2 using concentrated sulfuric acid and DI water. Before commencing with each leaching
process, a non-magnetic eudialyte concentrate was produced via the WHIMS to reduce the amount of iron
and analyzed in the ICP-MS.
Leaching Process 1
All experiments consisted of a one-liter solution of 0.1 M sulfuric acid, approximately 100 grams
of non-magnetic concentrate sample and an agitator to keep the sample suspended. Figure 3.14 shows the
38
|
Colorado School of Mines
|
experimental set-up for leaching process 1. Free acid titrations were also conducted after sample addition
and every half an hour after to assess free sulfuric acid concentration. The free acid titration used 5-mL
of leach liquor, 1-mL of 0.5 g/L methyl orange indicator solution and 1.0 N sodium carbonate titrate
solution. The titre volume was recorded. In experiment 1.1, the sample was added to a two-liter beaker
containing the acid solution. Immediate addition of the eudialyte sample increased the pH from 1.0 to
almost 3.0. The pH was brought back to 1.0 with the addition of concentrated sulfuric acid. This
leaching experiment lasted two hours. In experiment 1.2, the acid solution was heated to 50°C before the
addition of the eudialyte sample. Again, concentrated sulfuric acid was used to bring the pH value to 1.0
after the addition of the sample increased it. This experiment also lasted two hours. In experiment 1.3,
the acid solution was heated to 75°C before the addition of the eudialyte sample and concentrated sulfuric
acid was used to bring the pH value to 1.0 after the addition of the sample increased it. This experiment
also lasted two hours. Experiment 1.4 was similar to experiment 1.1, except it was done for four hours.
Experiments 1.5 and 1.6 were similar to experiments 1.2 and 1.3, respectively and were conducted for
four hours each. During the third hour of experiment 6, the leach liquor became viscous. Consequently,
more acid solution was added and the agitator rpm was adjusted. Once all the experiments were
completed, the leach liquor was vacuum filtered. The filtrate was prepared for ICP-MS analysis and the
filter cake was dried and weighed before XRF analysis.
Figure 3.14. Experimental set-up for leaching process 1.
39
|
Colorado School of Mines
|
Leaching Process 2
All experiments consisted of just enough 98% pure sulfuric acid addition to wet the sample,
approximately 25 grams of non-magnetic concentrate sample and 150-mL of DI water were used to
obtain the sulfates in solution. Figure 3.14 shows the acid-wet ore sample for leaching process 2. In
experiment 2.1, 20-mL of acid were added to the sample and was left to air dry. The DI water was added
once the sample dried. In experiment 2.2, 8-mL of acid were added to the sample and left for one hour.
After, it was placed in a furnace at 60°C to dry. Once dried, DI water was added. Finally, in experiment
2.3, 8.4-mL of acid was added and left to sit for 30 minutes before immediate DI water addition. The
leach liquors were then vacuum filtered. The filtrate was prepared for ICP-MS analysis and the filter cake
was dried and weighed before XRF analysis.
Figure 3.15. Experimental set-up for leaching process 2.
40
|
Colorado School of Mines
|
TECHNIQUES USED FOR ANALYSIS OF RESULTS
This chapter details the analytical techniques employed for the evaluation and interpretation of
the beneficiation and hydrometallurgy work done. MLA and TOF-SIMS was used for the analysis of the
received Norra Kӓrr eudialyte sample. The use of the XRF and ICP-MS was used for the beneficiation
and hydrometallurgical treatment product analysis.
4.1 Mineral Liberation Analysis (MLA)
One of the most widely used technologies for automated mineralogy in the mineral and metal
industries, MLA is an automatic mineral analysis system that identifies minerals in polished sections.
MLA also quantifies many mineral characteristics, such as abundance, grain size and liberation. This is
done by combining an automated Scanning Electron Microscope (SEM) and multiple Energy Dispersive
X-ray detectors. [75,76] This type of analysis is essential in better understanding the mineralogy for
optimizing a certain beneficiation and/or hydrometallurgical process specific to the mineral of interest.
The Norra Kärr eudialyte sample was sent to The Center for Advanced Mineral and Metallurgical
Processing (CAMP) at Montana Tech of the University of Montana for MLA
4.2 X-ray fluorescence (XRF)
XRF is a non-destructive analytical method for determining the elemental composition of
materials through fluorescent X-ray measurement. Fluorescent or secondary X-rays are emitted from a
sample when it is excited by a primary X-ray source. Each element has a characteristic fluorescent X-ray
“fingerprint” that it can be associated with; the energy of the electron depends on the shell it occupies and
the element it belongs to. [54] When an atom in the sample is hit with an X-ray of energy greater than the
atom’s K or L shell binding energy, an electron from one of the shells is displaced. To regain stability, an
electron from one of the higher energy shells fills the vacancy. When this electron drops to a lower
energy state, a fluorescent X-ray is emitted whose energy is equal to the difference in energy between the
two quantum states. [55]
A borate fusion was done to solubilize the mineral samples for the XRF. The oxidized sample
was dissolved in the molten flux at temperatures between 1020-1050°C with a 2:1 ratio of lithium
metaborate to lithium tetraborate due to the acidic nature of the mineral. The fusion was conducted in a
Katanax K1 Prime fluxer that produced a glass disk for XRF analysis. The glass disks are homogenized
to reduce particle size, mineralogy and matrix effects. [56]
A quantitative method was developed in the XRF using two analytical standards provided by
Tasman Metals, Ltd., and a rare earth mineral standard with a similar total rare earth element composition
as the eudialyte sample. The rare earth mineral standard was chosen based on its similar composition to
41
|
Colorado School of Mines
|
the eudialyte material tested. The value of silica and rare earth elements in the standard were the main
values of comparison. The certified values for the two analytical standards provided by Tasman and the
rare earth mineral provided by Brammer Standard Company, Inc., are shown below.
Table 4.1. Certified values for Brammer Standard Company rare earth mineral.
Compound Certified Value (%)
Σ REO 0.764
Al O 19.00
2 3
Fe O 3.46
2 3
K O 2.11
2
SiO 66.72
2
CeO 0.023
2
Dy O 0.021
2 3
Er O 0.011
2 3
Eu O 0.00750
2 3
Gd O 0.026
2 3
Ho O 0.0049
2 3
La O 0.277
2 3
Lu O 0.00136
2 3
Nd O 0.186
2 3
Pr O 0.054
2 3
Sc O 0.00118
2 3
Sm O 0.033
2 3
Yb O 0.0100
2 3
Y O 0.124
2 3
When reporting the results in the following sections for the beneficiation experiments, it is worth
noting that the TREE values include those rare earth elements listed above, which the XRF was
programmed to analyze for.
4.3 Inductively Coupled Plasma- Mass Spectrometry (ICP-MS)
ICP-MS is another analytical technique used for the determination of elemental composition. The
basic principle of the ICP-MS is to convert the atoms of a sample into ions that are separated and detected
by the mass spectrometer. Argon gas flows through the ICP torch and when ignited, electrons are
42
|
Colorado School of Mines
|
provides elemental and molecular information when the solid sample surface is bombarded with a pulsed
primary ion beam that is generated by a liquid-metal ion gun. [73] The ion beam causes molecular ions to
be dislodged from the sample surface and are directed to an analyzer for mass measurement based on
their time of flight to the detector. The basis of this analytical technique is that ions ejected from the
sample surface at the same energy will travel at different velocities due to their different mass, resulting in
lighter ions to arrive before heavier ions. [73] The Norra Kärr eudialyte sample was mounted in epoxy
and polished, with the goal being to detect Fe, Al, Si, Y, Ce, Zr and lanthanides over a 500-micron area.
[74]
Table 4.3. Certified values for NKA01 analytical standard.
Compound Certified Value (%)
Σ REO 0.40
Al 8.65
Fe 5.58
K 3.02
Si 25.17
Zr 1.29
(ppm)
Ce 626
Dy 175
Er 129
Eu 10.8
Gd 111
Ho 40.4
La 313
Lu 18.3
Nd 307
Pr 79
Sc < 1
Sm 88
Yb 128
Y 1131
44
|
Colorado School of Mines
|
RESULTS AND DISCUSSION
The results of the experiments detailed in sections 3.3 and 3.5 are discussed in this chapter.
Beneficiation test data is presented first, followed by the hydrometallurgical leaching test results.
5.1 Beneficiation Experimental Results
This section provides discussion on the results from the heavy liquid separation tests conducted
using sodium polytungstate, followed by those conducted using methylene iodide. Finally, the wet high
intensity magnetic separation test results are presented.
Sodium Polytungstate Heavy Liquid Separation Test
As discussed before, sodium polytungstate is a heavy liquid used instead of many traditionally
used heavy liquids due to its non-toxic behavior. Figure 5.1 shows the recovery of total rare earth
elements (TREEs) as a function of the specific gravity.
Recovery of TREEs as Specific Gravity Increases
100
90
)
% 80
(
y
r e 70
v
o
c e 60
R
50
40
2.7 2.95 3.08
Specific Gravity
100x140 140x200 200x270 270x400 (-400)
Figure 5.1. Recovery of total rare earth elements in five size fractions with increased specific gravity.
The results indicate that a maximum recovery of TREEs for each size fraction is achieved at a specific
gravity of 2.95. The recovery of TREEs at this specific gravity vary between 80 and 95%. Recovery
values above 90% are obtained for the 100x140, 140x200 and 200x270 size fractions which echoes the
previous claim made during the characterization of the eudialyte ore, that the most liberated size fraction
is 200x400 mesh. The recovery of TREEs shows to decrease for all size fractions at the 3.08 specific
gravity. At 3.08, we see the -400- mesh size fraction achieves about 45% recovery. These are points of
interest because it is assumed a greater liberation is achieved at -400-mesh allowing for better separation
45
|
Colorado School of Mines
|
of the aegirine from the eudialyte and other gangue minerals. The specific gravity of 3.08 is also the high
limit for the eudialyte mineral density, and the assumption is made that this specific gravity would
produce a better separation between aegirine and eudialyte, since it is in between the densities of both
minerals. However, both assumptions are proven incorrect since the -400-mesh size fraction does not
show a maximum of recovery above the other fractions, and the recovery decreases at the highest specific
gravity of 3.08.
Figure 5.2 shows the recovery of zirconium as a function of the specific gravity, for the five size
fractions. Zirconium is used as an indicator for the eudialyte mineral, along with the TREEs. The
recovery of zirconium is similar to that of the TREEs, which is as expected, since as the zirconium-heavy
eudialyte mineral is recovered, the rare earth elements are recovered as well. A maximum recovery of
zirconium is also achieved at a specific gravity of 2.95, along with a decrease in recovery at the higher
specific gravity of 3.08.
Recovery of Zr as Specific Gravity Increases
100
80
)
%
( 60
y
r
e
v
o 40
c
e
R
20
0
2.7 2.95 3.08
Specific Gravity
100x140 140x200 200x270 270x400 (-400)
Figure 5.2. Recovery of zirconium in five size fractions with increased specific gravity.
Tables 5.1-3 show elemental upgrade ratios at the specific gravities of 2.7, 2.95 and 3.08,
respectively, for the five size fractions. As described before, the purpose of this heavy liquid separation
test was to separate the iron-bearing aegirine mineral from the eudialyte and other gangue. Since the
aegirine is the heaviest of the gangue present, at all gravity values, it was expected to report to the sink
fraction, while the eudialyte and lighter gangue were to report to the float fraction. The upgrade ratio is
calculated with respect to the fraction it was supposed to report to. In each table below, the upgrade ratio
for the rare earth elements and zirconium stay consistently around 1.2, while upgrade ratio of iron seems
to vary between specific gravity values. The constant ratio for the TREEs and Zr indicate that there may
46
|
Colorado School of Mines
|
not be a dependence on specific gravity for upgrading eudialyte. However, for upgrading iron, the results
show some dependence on density, reaching an average maximum at the specific gravity of 2.95 as seen
in Table 5.2.
Table 5.1. Upgrade ratios for five size fractions at a specific gravity of 2.7.
Upgrade Ratio
TREEs Zr Fe
100x140 1.2 1.2 2.9
140x200 1.3 1.3 2.6
200x270 1.2 1.3 2.3
270x400 1.2 1.2 3.1
-400 1.1 1.2 3.1
Table 5.2. Upgrade ratios for five size fractions at a specific gravity of 2.95.
Upgrade Ratio
TREEs Zr Fe
100x140 1.2 1.2 3.4
140x200 1.2 1.2 3.0
200x270 1.2 1.2 2.9
270x400 1.1 1.1 3.6
-400 1.2 1.2 2.7
Table 5.3. Upgrade ratios for five size fractions at a specific gravity of 3.08.
Upgrade Ratio
TREEs Zr Fe
100x140 1.2 1.2 2.5
140x200 1.2 1.3 2.4
200x270 1.3 1.2 2.4
270x400 1.3 1.4 2.2
-400 1.0 1.1 1.7
In all, these recovery and upgrade ratios provide evidence for the use of zirconium as a eudialyte
and thus, rare earth element indicator, since the zirconium and total rare earth element values matched up.
However, the results do not demonstrate a significant upgrade of eudialyte via heavy liquid separation test
with sodium polytungstate.
47
|
Colorado School of Mines
|
Methylene Iodide Heavy Liquid Separation Test
Unlike sodium polytungstate, methylene iodide is considered a toxic heavy liquid and was used in
this experiment to reach a density of 3.2 g/cm3. Figure 5.3 shows the recovery of TREEs for four size
fractions at the specific gravity of 3.2. As the average particle size increases from 29 microns to +400-
mesh, the recovery of the TREEs decreases. The same trend is shown in Figure 5.7, where the recovery
of zirconium is shown for four size fractions at a specific gravity of 3.2. As described above in section
3.2, this heavy liquid separation test was done to investigate the differences between the as-received ore,
pulverized and screened samples. Figures 5.7-8 illustrate a greater recovery of TREEs and Zr for the
pulverized size fraction of 29 microns, indicating a certain degree of liberation is achieved when the as-
received ore sample is pulverized. This degree of liberation may be due to the change in surface
morphology by pulverizing the ore, where a single particle is crushed/ground into smaller particles, thus
revealing a new surface. This differs from screening the as-received ore sample because in screening, the
smaller particles are separated, not produced by crushing or grinding.
a) b)
c) d)
Figure 5.3. As-received sample: a) float product, b) sink product, c) magnetic fraction and d) non-
magnetic fraction.
Table 5.4 shows the elemental upgrade ratios for the four size fractions at a density of 3.2 g/cm3
for each size fraction. As can be seen by the upgrade ratios for each elemental group, there is not a lot of
difference in upgrade ratios for each size fraction. However, there is a greater upgrade ratio for iron in all
size fractions in this heavy liquid separation test.
48
|
Colorado School of Mines
|
Wet High Intensity Magnetic Separation on Float Products
WHIMS was done on the float products of the heavy liquid separation tests done using methylene
iodide. Figures 5.9 and 5.10 display the recovery of the TREEs and Zr for each size fraction,
respectively. It was expected for the recovery of the TREEs and Zr to be high in the magnetic fraction
since the float product consisted of the paramagnetic eudialyte and non-magnetic gangue. However, the
results show the TREEs and Zr reporting to the non-magnetic fraction. Table 5.6 shows the upgrade ratio
values for TREEs and Zr in magnetic fraction for each size fraction. Again, there is no significant
upgrade of the eudialyte ore, as indicated by these values. These values also do not show dependence on
particle size for better separation.
Recovery of TREEs
80
70
60
)
%50
(
y
r e40
v
o
c30
e
R
20
10
0
29 microns (-400) mesh 111 microns (+400) mesh
Magnetic Non-Magnetic
Figure 5.9. Recovery of total rare earth elements in magnetic fraction from float products at 1 Tesla.
Due to the complex and variable chemical composition of eudialyte, it may be suggested that this
Norra Kӓrr eudialyte ore does not exhibit paramagnetic behavior compared to other eudialyte minerals
studied in literature. However, before such a conclusion can be made regarding this specific eudialyte
sample, more magnetic separation tests should be conducted at higher magnetic strengths. Magnetic
separation of eudialyte may also be highly variable with respect to which fraction the mineral will report
to (magnetic or non-magnetic) due to the iron present.
52
|
Colorado School of Mines
|
Recovery of Zr
80
70
60
)
%50
(
y
r e40
v
o
c30
e
R
20
10
0
29 microns (-400) mesh 111 microns (+400) mesh
Magnetic Non-Magnetic
Figure 5.10. Recovery of zirconium in magnetic fraction from float products at 1 Tesla.
Table 5.6. Upgrade ratios for magnetic fraction on float product at 1 Tesla.
Upgrade Ratio
TREEs Zr
29 microns 1.1 1.0
-400 mesh 1.8 1.5
111 microns 1.4 1.2
+400 mesh 1.6 1.3
5.2 Hydrometallurgical Treatment Results
This section presents the data and discusses the results obtained by the experiments in each leach
process. The results for leach process 1, experiments 1.1–1.3 are discussed first, followed by experiments
1.4–1.6. Finally, leach process 2 results are examined.
Leach Process 1
A discussion regarding the experiments conducted for the first leaching process is presented in
this section. In experiments 1.1–1.3, the leach liquor was slow to filter and the filtrate leach solution
gelled. However, the rate at which these leach solutions filtered and gelled differed and may be
correlated to the temperature at which they were leached. The leach solution from experiment 1.3 gelled
first, two days after the test was conducted and was comparatively the slowest to filter. The leach
solution from experiment 1.2 gelled a week after the test. Finally, the leach solution from experiment 1.1
took almost three weeks to gel, but relatively the fastest to filter. Figure 5.11 shows the recovery of the
53
|
Colorado School of Mines
|
The figure below displays the free acidity values in grams per liter of sulfuric acid for experiments 1.1–
1.3 and shows a large decrease after the initial titration done. For the remainder of the leaching
experiments, the free acidity stays relatively constant. Table 5.7 shows how much acid is consumed
during each experiment. Experiment 1.1, conducted at 25°C shows the highest amount of acid
consumption at 7.7 g, while the other experiments, conducted at higher temperatures show lower acid
consumption. The initial high free acidity value followed by the decrease corresponds to the increase in
pH with the addition of the eudialyte sample. As acid was consumed in the experiment, the pH value rose
from 1 to 3.
Free Acidity at 2 hour leach
10.0
8.0
)
L
/g
( 6.0
d
ic
A
4.0
e
e
r
F
2.0
0.0
30 60 90 120
Time (minutes)
Experiment 1.1 Experiment 1.2 Experiment 1.3
Figure 5.13. Free Acidity for sulfuric acid for leaching experiments at two hours.
Table 5.7. Consumption of sulfuric acid for leaching experiments at two hours.
Experiment Acid Consumed (g)
Experiment 1.1 7.7
Experiment 1.2 3.5
Experiment 1.3 3.5
Also in experiments 1.4–1.6, the leach liquor was slow to filter and the filtrate leach solution
gelled. Similar to the experiments conducted for two hours, the rate at which these leach solutions
filtered and gelled differed and may be correlated to the temperature at which they were leached.
Extreme gelation occurred during experiment 1.6, where the leach liquor thickened after three hours at
75°C. For the last hour of the experiment, the agitator rpm was increased. In comparison to experiments
55
|
Colorado School of Mines
|
1.1–1.5, this leach liquor was the most difficult and slowest to filter, resulting in a relatively small amount
of filtrate. The gelation formed in experiment 1.6 caused a large hydration of the slurry and product,
resulting in high recovery values. Figure 5.14 shows a recovery value above 120% for the total rare earth
elements, while the recovery value for zirconium has been omitted from Figure 5.15 due to its
unreasonable value. Figure 5.15 also does not show an R2 value because two data points is not sufficient
for a linear relationship. A relationship between temperature and recovery of TREEs and Zr is a little
more difficult to present in this set of four-hour experiments due to the gelation, however, an upward
trend is present.
Figure 5.16 displays the free acidity in grams per liter of sulfuric acid for experiments 1.4–1.6,
where a decrease in free acid is shown, similar to experiments 1.1–1.3. The free acidity values for
experiment 1.6 do not remain constant, but instead show some deviation from a constant trendline. This
again, may be due to the thick gelled formed during the experiment. Table 5.8 shows the acid
consumption data for each experiment. Like the previous three experiments, experiment 1.4, conducted at
25°C, shows the highest acid consumption.
Recovery of TREEs vs. Temperature
Experiment 1.6
130
R² = 0.8503
120
110
)
%100
(
y Experiment 1.5
r e 90 Experiment 1.4
v
o
c 80
e
R
70
60
50
20 30 40 50 60 70 80
Temperature (°C)
Figure 5.14. Recovery of total rare earth elements for leaching tests conducted at four hours and three
different temperatures.
56
|
Colorado School of Mines
|
The figures shown above display show a linear relationship between temperature and recovery of
the total rare earth elements and zirconium. While temperature is an important parameter, the leaching
time is an equally important variable to consider. Figures 5.17 and 5.18 show the recovery of TREEs and
Zr at each temperature for both time intervals. Taking into consideration the unreasonable TREEs and Zr
recovery values due to gelation, there is an increase in recovery as leaching time increases. Increased
leaching time allows the acid to permeate more particle surfaces, since the ore is in solution for a longer
period of time. However, at an increased temperature, such as in experiment 1.6, this increased exposure
to the acid and water will lead to gel formation.
Recovery of TREEs
140
120
)100
%
(
y 80
r
e
v 60
o
c
e
R 40
20
0
25 50 75
Temperature (°C)
2 hours 4 hours
Figure 5.17. Recovery of total rare earth elements as a function of time and temperature for all
experiments.
As discussed above, in all of the leaching experiments conducted under process 1, the leach solutions
eventually gelled respective to their kinetics. Figure 5.19 shows a sample of the filtered leach solution
after it has been left alone for a few days and gelled. As can be seen from Figure 5.19, the gelled mass is
transparent since the solution was filtered. It can be easily broken apart with minimal force and has a
hydrated consistency as is expected with a gel.
Leach Process 2
A discussion regarding the results of the second leaching process is presented in this section.
Figures 5.20 and 5.21 show the recovery of the total rare earth elements and zirconium as a function of
retention time in the sulfuric acid, respectively. In each experiment, the exposure time to the acid and
leaching water was varied.
58
|
Colorado School of Mines
|
Recovery of Zr
100
80
)
%
( 60
y
r
e
v
o 40
c
e
R
20
0
25 50 75
Temperature (°C)
2 hours 4 hours
Figure 5.18. Recovery of zirconium as a function of time and temperature for all experiments.
Figure 5.19. Gelled leach solution.
The sample in experiment 2.3 did not experience any drying time since the DI water was immediately
added after the 30- minute retention time in acid. Upon the addition of the DI water, the solution became
cloudy and bubbled. The sample in experiment 2.2 was oven-dried after a one-hour retention time in the
sulfuric acid. The addition of DI water to the oven-dried sample did not immediately bubble like in
experiment 2.3, but some bubbles were observed as the solution was left sitting for about ten minutes. In
experiment 2.1, the acid-wet sample was left to air dry for 24 hours before DI water was added. Similar
to the oven-dried sample, this solution did not immediately bubble until it after about ten minutes. While
neither of these experiments showed gelation during the acidification or after the DI water was added, nor
did the filtrate solutions gel, the addition of water did cause a reaction to occur, evident by the bubbling.
59
|
Colorado School of Mines
|
PRELIMINARY ECONOMIC ANALYSIS
This chapter includes a preliminary economic analysis for this project to discuss the economic
viability of the proposed mineral processing and leaching treatment conducted on an industrial level.
The proposed economic model is taken from a basic CostMine model and transformed to fit
operations needed for this project. Zirconia price values are included in addition to the rare earth
elements in this analysis since REE grade is not very high and both can essentially be separated through
hydrometallurgical techniques. Initially, an economic analysis looking solely the REE recovery was done
and yielded a higher operating cost than the profit that would be made per year. The advantage of
recovering the zirconium with the rare earths in the leaching step results in a greater profit. Tables 6.1a-c
show the initial production, operating and capital costs, respectively.
Table 6.1a. Production costs of recovery of rare earth elements and zirconium in a 2,000 tonnes per day
process.
Production Day Production Year Production
Ore Mined 2,000 tonnes/day 730,000 tonnes/year
REE Grade (%) 0.58 0.58
REE Recovery (%) 87.0 87.0
Tonnes of REE (tonnes) 5 1,770
REE Price ($/tonne) 4,000.00 4,000.00
Zr Grade (%) 0.96 0.96
Zr Recovery (%) 75.0 75.0
Tonnes of Zr (tonnes) 7 2,509
ZrO Price ($/tonne) 15,500.00 15,500.00
2
Working Hours 24 hours/day 24 hours/day
Working Days 365 days/year 365 days/year
Schedule 8,400 hours/year 8,400 hours/year
Total Production Costs (USD) 274,949.49/day 96,232,320.00/year
This economic analysis starts with the basis of processing 2,000 tonnes of ore per day, resulting in
730,000 tonnes of ore per year, as can be seen in Table 6.1a. This rate yields a rare earth element (REE)
production of 5 tonnes per day and 1,770 tonnes per year, and a zirconium production of 7 tonnes per day
and 2,509 tonnes per year. Production costs are calculated on the assumption of a 24-hour work day, 365-
days a year. Mining costs were based on a surface mine with a strip ratio of 1:1 (waste: ore).
62
|
Colorado School of Mines
|
CONCLUSION AND FUTURE WORK RECOMMENDATIONS
The goal of this research project was to investigate an efficient beneficiation and leaching method
to process the Norra Kärr eudialyte sample. An extensive literature survey was done on the eudialyte
group minerals, various beneficiation techniques and iron and silica gel issues in the leaching of silicate
minerals. Based on the literature review and the mineralogical characteristics studied on the Norra Kärr
eudialyte mineral, it was concluded that the mineral would be concentrated through a combination of
gravity separation followed by magnetic separation with a WHIMS, with the prospect of conducting a
leaching study. Preliminary heavy liquid separation tests followed by the WHIMS did not result in
significant upgrade ratios for the Norra Kärr eudialyte sample, with some experimental data suggesting
that this mineral sample is not paramagnetic or must be processed at a much higher magnetic field
strength to achieve better separation.
Focus then shifted to the hydrometallurgical treatment of the sample, more specifically, the
leaching experiments. Again, based on previous research conducted on other eudialyte minerals and
silicate minerals, two different leaching processes were developed, with the goal to recover as much of
the rare earth elements and zirconium. A preconcentrate would be produced via the WHIMS to remove
iron without removing the rare earth elements. The decision to create this low-iron preconcentrate for the
leaching experiments was made based on previous studies resulting in hindered hydrometallurgical
recovery due to iron in the concentrate. A preconcentrate was created and used for the leaching
experiments which resulted in 87% recovery of the rare earth elements and 75% recovery of zirconium.
Although it has been studied that a water starved system is best when leaching silicate minerals, the high
recoveries were obtained in experiments were the preconcentrate had access to excess water. While some
experiments did not show silica gel formation during the actual experiment, the filtered leach solution did
end up gelling according to the time and temperature it was conducted at. Experiments done at higher
temperatures for longer periods of time showed a faster gelation time after filtering compared to those
done at lower temperatures for a shorter amount of time.
The preliminary economic analysis showed a profit for the production of the rare earth elements
and zirconium with a payback period around 37 months. The zirconium production was included in this
economic analysis because it can be produced hydrometallurgically speaking, and will cover the operating
costs.
In all, while this combination of beneficiation techniques did not provide significant upgrading
results, eudialyte should still be considered a viable but still mineralogically unknown source for rare
earth elements. It is evident from previous research and the research conducted in this project that a
67
|
Colorado School of Mines
|
ABSTRACT
Gravity separation and flotation studies have been conducted on Molycorp
bastnaesite ore in order to determine if new beneficiation schemes present a more
selective and more economical alternative than that which is currently employed at
Mountain Pass. Literature on bastnaesite, monazite, barite, and calcite flotation and
gravity concentration principles was surveyed. Flotation reagent additions were
determined using components that have shown preferential floatability of bastnaesite
and monazite over the gangue minerals. Hallimond Tube microflotation tests were
performed on crushed and ground ore samples. Heavy liquid separation with sodium
polytungstate was used to investigate the effectiveness of gravity separation on the ore.
Shaking table and Falcon concentrator tests were performed to gravity concentrate the
ore. A gravity-concentrated feed was floated and compared with a non-concentrated ore
feed to illustrate the benefit of preconcentration. An economic analysis was generated
for flotation plants operating with and without gravity preconcentration that would sell
products with two distinct grades and recoveries.
Qualitative microflotation tests produced little selective separation of the rare
earth minerals (bastnaesite, parisite, and monazite) from the gangue (calcite, barite,
dolomite, and quartz). Heavy liquid tests illustrated the sink/float behavior of the
minerals at different specific gravities of separation. Their results suggest that at higher
specific gravities the calcite floats while the bastnaesite and barite sink. Shaking table
tests showed potential to effect such a separation, but optimum conditions were not
determined. A Falcon centrifugal concentrator was used to carry out tests according to a
Design of Experiments matrix generated with Stat Ease Design Expert 9. The best
conditions from those trials were determined, and the tests were repeated to verify the
desirability of those parameters. Bench flotation was then used to compare the standard
feed at plant conditions to a feed consisting of the blended gravity concentrates. The
flotation results showed that the preconcentrated feed outperformed the typical plant
feed. Economic analysis of a plant with and without gravity preconcentration shows that
gravity preconcentration, although more capital-intensive, will yield a higher annual
profit and a better 10-year net present value.
iii
|
Colorado School of Mines
|
CHAPTER 1: INTRODUCTION
In comparison to the broad spectrum of applications based on the rare earth
elements, their supply shows little diversity. Their applications range from polishing
media to hard drive magnets to water treatment additives to wind turbine motors. Prior
to the 1950s the mineral providing the bulk of the rare earth supply was monazite; a
phosphate mineral which was beneficiated primarily from placer deposits. There it could
be separated from associated ilmenite, garnet, magnetite, quartz, rutile, and xenotime
using a simple physical separation scheme utilizing differences in specific gravity,
magnetism, and conductivity. For years this was the source material for cerium,
lanthanum, neodymium, and thorium.[1]
In the 1950s the Mountain Pass Mine was developed by the Molybdenum
Corporation of America, now Molycorp Inc. This deposit in southern California is the
world’s richest source of bastnaesite, a fluorocarbonate containing cerium and
lanthanum, as well as the heavy rare earths neodymium and praseodymium. Separation
is much more difficult than with monazite from beach sands. This deposit contains
bastnaesite and monazite as the primary valuable minerals (10% of the ore) with calcite
and other carbonates (60%), barite (20%), and quartz and other minerals (10%).
Beneficiation of the rare earths from the bastnaesite ore at Mountain Pass
involves concentration by flotation followed by roasting, leaching with hydrochloric acid,
and solvent extraction. [1] Traditionally, Molycorp employed a crush, grind, float system
to produce a 63% rare earth oxide (REO) concentrate, shown in Figure 1.1. The
flotation conditioning involves alternating additions of steam, soda ash, lignin sulfonate,
and tall oil, shown stepwise in Table 1.1. Flotation occurs at temperatures near 82 °C in
several roughing, cleaning, and scavenging stages (Figure 1.2). The rougher produces
a 30% REO concentrate which is further upgraded to 63% REO through cleaning. This
is achieved at a 65-70% recovery.
1
|
Colorado School of Mines
|
China displaced the United States as the dominant supplier of rare earths in the
mid-1980s due to production from the Bayan Obo mine in Inner Mongolia, and now is
responsible for more than 90% of the world’s rare earths supply. Little effort has been
put forth toward seeking a domestic source of the elements due to the cheap price of
rare earths exported from China. Mountain Pass had been unable to compete with
Chinese producers; it halted production in 2002 and resumed stockpile processing in
2008.
Hendrick compiled a report on U.S. rare earth commercial activity in 2008. No
rare earths were mined, but stockpiled concentrate was processed at the Mountain
Pass Mine. Bastnaesite concentrate and monazite prices, respectively compiled from
USGS sources and U.S. import values, were $8,000/ton and $480/ton. [4] Again
straining the supply in the US was the 2010 decision by the Chinese government to limit
the volume of rare earth exports. This sanction caused a drastic spike in the price of the
rare earth elements (shown in Figure 1.3) and brought questions to light of their
availability, considering their importance in strategic applications.
Figure 1.3. REO Export Prices from China. Data retrieved from www.metal-pages.com
In 2011, the US Department of Energy released its Critical Materials Strategy;
identifying rare earth materials as “critical” and in need of a more reliable domestic
4
|
Colorado School of Mines
|
supply. From this, the Critical Materials Institute was created with Molycorp as a partner
in order to investigate their complex beneficiation problem. [5] Gleason’s 2011 article
highlighted Molycorp’s strategic plan to resume production of rare earth minerals at
Mountain Pass up to a goal of 40,000 metric tons/year by 2013. It also discussed some
of the policy proposals laid out to accelerate the United States’ rare earth metals
production. [6]
As of February 2015, Molycorp struggled to upgrade plant operations and
compete with low-cost rare earth products from China. Rare earth oxide equivalent
production by Molycorp for 2014 was 4,785 mt, which was an increase from 3,473 mt in
2013. [7] Rare earth prices had fallen dramatically from their highs in 2011. Due to this,
Molycorp was forced to suspend operations in October of 2015 and operate only for
care and maintenance of the operation. [8]
The general forecast is that an increase in rare earth production is needed to
meet growing worldwide demand, and it will not be met from Chinese supply alone.
More than 90% of the world’s rare earths come from China, but that comes from only
25% of the world’s reserves, accounting for recent estimates containing Canadian and
Australian reserves.
Beneficiation of rare earths was recently summarized by Jordens, focusing on
bastnaesite at Mountain Pass and Bayan Obo and monazite beach sands elsewhere in
the world. Extensive Chinese knowledge in rare earth element (REE) flotation abounds
in papers but many of them lack scientific depth and/or accuracy. Gravity concentration
of rare earth minerals has been historically difficult due to the influence (and loss) of
rare-earth-containing fines and the similar specific gravity of barite. [9] Limited
commercial success has been seen with gravity separation of bastnaesite. The Sichuan
Mianning REE Ore deposit uses shaking tables to separate minerals from pre-classified
feeds. The mineralogy there is a carbonate containing barite, fluorite, and iron- and
manganese-containing minerals with a 3.7% REO grade – most of which is bastnaesite
– in coarse (>1mm) and fine powder (80% -325 mesh Tyler ) sizes. An all-gravity
operation classifies and concentrates a 62% passing 200 mesh feed to achieve grades
of 30%, 50%, and 60% with an overall recovery of 75%. Gravity and flotation are also
5
|
Colorado School of Mines
|
combined (Figure 1.4) to concentrate the feed and produce an overall 30% grade
gravity concentrate at 74.5% REO recovery. The concentrate is then reground to 70% -
200 mesh and floated with hydroxamic acid, phthalate, sodium carbonate, and sodium
silicate to produce a 50-60% REO grade concentrate with a rare earth recovery of 50-
60%. [10] Lab scale beneficiation examples are more abundant. A Mozley Multi-Gravity
Separator was used to separate a Turkish bastnaesite ore, producing a 35.5% REO
preconcentrate at 48% recovery. [11] Egyptian beach monazite was concentrated using
electrostatic, magnetic, and gravitational methods. [12] The beach sands were screened
to pass 1mm and deslimed then concentrated successively with a shaking table. The
concentrate was subjected to low intensity magnetic separation and the nonmagnetic
fraction beneficiated with a shaking table. The shaking table concentrate was dried and
processed using high tension electrostatic separation, magnetic separation, and a
shaking table again to produce a crude 85% monazite concentrate. More electrostatic
and magnetic separation produces a 97% monazite concentrate (Figure 1.5). Humphrey
spirals were used to concentrate an Iranian monazite ore. [13] Optimum results were
found with an intermediate feed size, high feed rate (1.5 L/s), and low solids density
(15%), with the latter parameter showing a less-significant effect. The best total rare
earth elements (TREE) grade was reported at 6050x10-6 percent with a 57.06%
recovery after gravity separation and leaching.
A developing site, Bear Lodge, owned by Rare Element Resources (RER) has
achieved success with purely physical concentration. The deposit is a carbonatite,
containing rare earths mostly as bastnaesite, with three regions of mineralization: oxide,
sulfide, and transition. Drill core test work showed that the oxide core was successfully
treated by scrubbing and sizing. The rare earth oxides were found to concentrate in the
passing 500 mesh size. Results of scrubbing tests on the ore are shown in Table 1.2,
where it can be seen that a 10-minute scrub yielded the maximized results in relation to
recovery, weight rejected, and scrub time. [14]
6
|
Colorado School of Mines
|
CHAPTER 2: LITERATURE REVIEW
Beneficiation of rare earth minerals has been reviewed previously. [15], [16]
Special attention is given to the minerals (bastnaesite, monazite, xenotime, calcite, and
barite), the collectors (oleic and hydroxamic acids), and the modifiers (soda ash, lignin
sulfonate, metal salts, sodium silicate) used in this study. Methods of gravity
concentration have also been surveyed. Prior to the discussion of reagents and their
effect in flotation systems, a review of surface chemistry and flotation phenomena, as
well as common analysis techniques, will be presented.
2.1 Flotation Surface Chemistry and Analysis Techniques
Flotation is a mineral processing technique that is widely used to concentrate an
ore before recovery by pyrometallurgical or hydrometallurgical routes. It involves
bubbling air through a tank of ground and crushed mineral pulp. Reagents are
specifically chosen so as to cause selective adsorption of the desired mineral to the air
bubbles. The mineral then rises through the pulp attached to the bubble where it sticks
to the other bubbles in the froth. The froth is skimmed off or collected via overflow, and
contains the upgraded ore. Often several steps are needed after concentration in the
initial cell (“roughing”) to recover value from the concentrate (“cleaning”) or tails
(“scavenging”). Several phenomena are at play, including the hydrophobicity of the
minerals, the electrical potential of the minerals and solution, dissolution of mineral
species, exposed surface area, and adsorption of reagents to mineral surfaces; many of
which are affected by the temperature. Laboratory testing of flotation systems occurs in
specially designed cells originally developed by A.F. Hallimond.
Fuerstenau collected data on contact angles, adsorption density, zeta potential,
and flotation rate for a quartz-dodecylamine system over the full pH range. By plotting
them together on sensible scales, their correlation was shown. That work refuted the
claim that solid-liquid interface phenomena (adsorption density, zeta potential) cannot
be connected to solid-liquid-gas interface phenomena (contact angle, flotation rate). [17]
Hence many of these techniques can be used to predict overall flotation behavior. The
9
|
Colorado School of Mines
|
benefits and shortcomings of some of the industry’s common floatability tests have been
previously detailed. There is no one industry standard, because all of the tests seem to
have some sort of bias, whether it is operator expertise, non-industrial chemistry,
unrealistic conditions, intensive time requirements, or some combination thereof. [18]
2.1.1 Hydrophobicity and Contact Angle
Whether or not water attaches to a surface (of a mineral in this case) is qualified
by its hydrophobicity, it’s “fear of water”. Water will attach to and wet a hydrophilic
surface, but not attach to a hydrophobic substance. Sulfur and graphite are very
hydrophobic, while calcite, quartz, and gypsum are hydrophilic materials. The source of
this attraction is in the interfacial energies of the solid/air, solid/liquid, and liquid/air
interfaces, related by Young’s Equation, represented visually in Figure 2.1:
Figure 2.1. Visual Representation of Young's Equation, showing the contact angle of a
hydrophobic (left) and hydrophilic (right) surface. [19]
The contact angle can be experimentally determined using the sessile drop
technique, which involves taking high-framerate consecutive images of a water drop as
it is placed on a mineral surface. The geometry used to determine the contact angle is
set up Figure 2.2. At the moment it hits the mineral surface, the water droplet forms a
dome with a base diameter of d. If this dome were to be extended below the surface to
form a sphere, that sphere’s radius would be R. The height of the dome above the
mineral surface is given as a, and b is the difference in length between the sphere’s
radius and the dome’s height (R minus a). The contact angle is then found at the
tangent point T on the mineral surface using the triangle OBT.
10
|
Colorado School of Mines
|
O T
B
Figure 2.2. Illustration of the geometry used to determine the contact angle of water on
carbon. [19]
Studies on bastnaesite, monazite, calcite and barite have shown that, like most
minerals, they have hydrophilic surfaces. Flotation is a matter of selectively making
mineral surfaces hydrophobic so they will become aerophilic, bind to the air bubbles,
and float.
2.1.2 Hydrolysis Reactions
When an ionic compound is placed in water, it will dissolve until equilibrium is
reached. When sodium chloride does this, it leaves effectively no solid behind as the
sodium and chlorine ions diffuse away from one another surrounded by shells of
protons or electron pairs from water molecules. The result is a neutral solution. Some
compounds, such as weak acids, equilibrate with a large concentration of the neutral
species. The family of weak acids includes fatty acid collectors, like oleic acid. As it
dissolves, a proton is taken away from the neutral acid and what remains is a charged
molecule that can adsorb to an oppositely-charged mineral surface. Oleic and alkyl-
hydroxamic acids can form salts with sodium and potassium to become sodium oleate
and potassium octyl hydroxamate (when the alkyl chain is an octyl group). Collector
structures and features will be detailed later. Another hydrolysis phenomenon takes
place at the mineral surface, where exposed ions can attract charge-balancing H+ or
OH- ions. The hydrolysis of cerium is particularly relevant in rare earth flotation: the
11
|
Colorado School of Mines
|
2.1.3 Zeta Potential and Point of Zero Charge
For salt-type minerals, unequal dissolution can occur because the ionic
components of the salt have different sizes and charges, which do not enter solution at
the same rate. Because of this mineral surfaces are charged in solution. That surface
attracts oppositely-charged ions from the bulk solution and creates an electrical
potential. The ions manipulating the surface charge are called the potential-determining
ions, and can consist of H+, OH-, CO 2-, SO 2-, and other ions (particularly those
3 4
dissolving from the mineral). The Stern plane, illustrated in Figure 2.4, is the name given
to the plane at which bound counter ions cannot come closer to the surface. The shear
plane is the plane at which ions are capable of motion in the solution when forced. The
potential at the Stern plane cannot be determined experimentally, but the potential at
the shear plane can be, and is known as the zeta potential. At a specific pH this
potential becomes zero. That pH is referred to as the PZC, the point of zero charge, for
that mineral.
Figure 2.4. Schematic of electrical double layer (from Somasundaran 1975). [23]
13
|
Colorado School of Mines
|
Figure 2.6. Zeta potential of barite, bastnaesite, and calcite in pure water. (Smith 1986,
from [16]).
It has been determined that in the pH window for flotation, calcite is positively
charged, bastnaesite is negatively charged, and the barite surface is undergoing
transformation to a carbonate surface. [22]
Reagents are responsible for changes in the PZC of minerals during flotation by
adhering to their surfaces and thus changing their charge. Bastnaesite, barite, and
calcite surface chemistry has been analyzed in response to changes in soda ash
concentration (used to modify the pH). The carbonate ion had the most pronounced
effect on the bastnaesite zeta potential, and the least on the calcite potential. The barite
zeta potential changed dramatically from positive to negative at 8x10-4 M carbonate due
to the formation of barium carbonate on the mineral surface. [25] Initially, the zeta
potential of the bastnaesite is more negative with respect to barite, but the barite
becomes more negative after an addition of 1x10-4 kmol/m3 ammonium lignin sulfonate,
because of the stronger adsorption onto the barite surface. [26]
Cheng’s PZC of monazite was reported as a pH of 5.3. This, combined with the
negative zeta potential at high pH, leads to the conclusion of oleate ions being
chemisorbed onto the surface. [27] The monazite zeta potential curves are shifted to the
left (the PZC becomes lower) and made steeper in the presence of hydroxamate
collectors. [28] Cheng also discussed computer modeling of CePO and YPO as the
4 4
16
|
Colorado School of Mines
|
basis for the wide range of reported PZC values (pH 3-7) for monazite and xenotime,
respectively. An alternative cause for variation was given: impurities from other ore
bodies and crystallinity, which determines which ions are exposed. [29]
2.1.4 Adsorption Density
Surfactants can bind to mineral surfaces in different configurations. The
mechanisms for surface attachment can be classified as low- or high-energy processes.
Physical adsorption involves van der Waals bonding and hydrogen bonding, while the
higher energy chemical adsorption relies upon covalent bonding. Molecules can
attach themselves horizontally, leading to lower adsorption densities; or in vertical
configurations, leading to higher adsorption densities. Multiple layers can form as the
molecules match hydrophobic ends or hydrogen bond from one chain to another. As
many as six hydroxamate layers have been reported to form at the interface. [22]
Adsorption densities on barite (as a horizontal monolayer) and on calcite (as a
horizontal layer, then a vertical layer) are much lower than on bastnaesite. Calcite
exhibits the curious characteristic of a linear increase in adsorption perhaps, but
unconfirmed, as the result of calcium hydroxamate precipitation.
Figure 2.7. Oleate adsorption onto bastnaesite as a function of pH with and without pre-
boiling. (Smith 1986 from [16]
17
|
Colorado School of Mines
|
bastnaesite. Bastnaesite’s trivalent state (as opposed to the divalent state of the
alkaline-earths) is proposed as a factor in the increased adsorption. [20]
FTIR has been used to distinguish whether physical adsorption or chemisorption
occurs between monazite and bastnaesite and sodium oleate and potassium octyl
hydroxamate. Sodium oleate physically adsorbs onto bastnaesite and monazite at pH 9
and pH 3 and 8, respectively. Potassium octyl hydroxamate adsorbs chemically onto
both minerals at pH 9.3 and 9. FTIR was unable to distinguish whether or not
chemisorption at pH 8 occurs for monazite and sodium oleate. [30]
2.1.5 Hallimond Tube Flotation
Fuerstenau modified the design of the original Hallimond Tube to a version
similar to that used in this study. [31] It is one of the most common devices used for
microflotation tests, although due to its simplified design, it is not considered
representative of industrial flotation. [18] Several reasons for this are incomplete
chemistry (a frother is not necessary, difficulty in generating reliable grade-recovery
curves, and unrealistic flow conditions.
Although operation of a Hallimond tube is simple, the parameters are not
standardized. Gas composition and flow rate, stirring speed, specific water volume and
temperature, reagent additions, and flotation time are all determined by the researcher.
The tube is filled with a very dilute slurry (one gram of mineral per 100 mL of water), and
gas is bubbled through a frit at the bottom. A magnetic stir bar disperses the bubbles,
which attach to particles in the slurry. As the bubbles rise to the top of the water, they
break, and the mineral falls into the concentrate stem.
2.1.6 Temperature Effects
The literature shows that for all three minerals, collection increases with
increased temperature (Figure 2.11). Bastnaesite shows a more pronounced effect.
This is the impetus for Molycorp’s steam conditioning: increasing selectivity. The
endothermic nature of adsorption revealed that it is a chemical adsorption process and
temperature was proposed as a driver of the increased adsorption. [32] Flotation with
elevated conditioning temperatures has shown (Figure 2.12) that between hydroxamate
19
|
Colorado School of Mines
|
As the predominant supplying mineral of rare earth elements in the world its
beneficiation has been extensively studied and reviewed. [3], [9], [22] Pradip performed
electrokinetic, Hallimond tube and Denver cell flotation, adsorption, and x-ray studies in
his thesis. The surface chemistry of bastnaesite, barite, and calcite has been analyzed.
Flotation experiments have been conducted with fatty acids and hydroxamic acids,
along with additions of inorganic salts, organic ions and molecules, soda ash, and lignin
sulfonate.
2.2.2 Monazite
The initial source of rare earth elements, monazite, is a rare-earth phosphate
(La,Ce)PO . It was originally beneficiated as a nuclear reactor material from placer
4
deposits using gravity, magnetic, and electronic separation techniques. When
bastnaesite containing much lower amounts of thorium was discovered, monazite fell
out of fashion. It is found in deposits containing bastnaesite, and thus necessitates a
separation step for those ores. Monazite’s specific gravity varies from 5.0 to 5.4.
Pavez and Peres tested species of monazite, zircon, and rutile using three
different collectors (sodium oleate, potassium octyl hydroxamate, and a commercial
hydroxamate) and a depressant (sodium metasilicate). [28]
2.2.3 Calcite and Barite
Calcite is a carbonate mineral of calcium, CaCO . It is a semisoluble salt-type
3
mineral, and has been the subject of numerous investigations. [33], [34], [21], [35], [25]
It is present in Mountain Pass ore as one of the primary gangue constituents. It has a
specific gravity of 2.7.
The other major gangue mineral is the sulfate barite, BaSO . It has also been
4
studied, often in attempt to depress it from bastnaesite. [21], [22], [25], [26] Barite has a
specific gravity of 4.5.
22
|
Colorado School of Mines
|
2.3 Reagents
Industrial flotation requires the use of surface-modifying chemicals in order to
efficiently separate the desired minerals from the gangue. Collectors are those used to
target the desired minerals, while modifiers and depressants are used either to
promote flotation of the desired mineral or to inhibit gangue flotation.
2.3.1 Collectors
Mineral flotation relies on collectors to attach to desired mineral surfaces and
bubbles. In sulfide flotation, the dominant collector type are xanthates. In rare earth
flotation, the traditional molecules used are fatty acids, while hydroxamates are a
promising group undergoing laboratory study, with limited industrial application.
2.3.1.1 Fatty Acids
A fatty acid is an organic acid consisting of a chain of singly or doubly bonded
carbon atoms and a functional group capable of donating a proton. As fatty acid chain
length increases from 8 to 12 carbons, collector concentration required for flotation
decreases (Figure 2.13). [34] As shown in Figure 2.14, at a concentration of 10-3 mol/L
fatty acid, chain lengths of 11 and 12 carbons float well from pH 6-12.5. Smaller chains
show minimal recovery above pH 10. The proposed mechanism for the collector
adsorption to surface calcium and carbonate is by the reaction CaCO 3 + 2(RCOO-)
Ca(RCOO) + CO 2-.
2 3
The structure of oleic acid is shown in Figure 2.15. It is a monounsaturated 18-
carbon chain ending in a carboxyl group (a double-bonded oxygen and a hydroxyl group
are bonded to the end carbon). Flotation of barite, calcite, and fluorite was studied using
oleic acid and sodium oleate (the sodium salt formed by oleic acid and sodium ions).
Electrokinetic, Hallimond tube, and abstraction tests suggest that a layer of calcium
oleate forms around the minerals, which prevents further dissolution to equilibrium. Due
to their similar characteristics, selectivity between the three minerals was proposed as
unlikely to be obtained. [35]
23
|
Colorado School of Mines
|
Figure 2.16. Bastnaesite recovery by sodium oleate as a function of pH. [26]
Recovery of bastnaesite with and without additions of a depressant is shown in
Figure 2.16. The maximum is in the slightly-alkaline pH region. Bastnaesite, barite, and
calcite all float best around pH of 9.5 but bastnaesite flotation requires a lower
concentration of sodium oleate (3x10-4). By combining that fact with the calcite minimum
around pH 8.5, conditions for effective separation have been determined, although plant
practice dictates the use of depressants. [22]
The maximum floatability of monazite was shown to coincide with the maximum
concentration of Ce(OH)2+ and La(OH)2+ (from thermodynamic modeling), the greatest
particle-bubble adhesion, and a pH of 8.5 – 9. [27]
Sodium oleate floatability experienced a minimum between pH of 4-5. Maxima
exist on either side for monazite (3 and 7), zircon (3, 6-8), and rutile (3, 7-8).
Hydroxamate floatability occurs between 3 and 7, 2.5- 9, and 3-8, for monazite, zircon,
and rutile. [28] Shown in Figure 2.17, the percentage of calcite floated was 20% at 3x10-
6 M, 80% at 6x10-6M, and 95% at 10-5 M oleate concentration. [21]
25
|
Colorado School of Mines
|
Figure 2.17. Oleate concentration required to float salt-type minerals.[21]
2.3.1.2 Hydroxamates
Hydroxamate collectors are chelating molecules that contain several active sites
to which an ion can bond and a sufficiently long chain to provide the necessary
hydrophobicity. That chain (represented by R in Figure 23) can be an alkyl group with 7-
14 carbons, a naphthalene ring, or other organic group. The number of active sites
varies from specific molecule to molecule, but they all function by forming coordinating
bonds with metal ions. [36] The specific mechanism of attachment can also vary but an
example is given in Figure 2.18, where the hydroxyl group is deprotonated and the two
oxygens coordinate to bind the metal ion. Different layer formations (either monolayer or
multilayer) are the result of horizontally versus vertically oriented molecules. These
layers form due to chemisorption with surface cations: cations form hydroxy complexes,
readsorb, then bond with the hydroxamate. [20]
Figure 2.18. A mechanism of hydroxamate adsorption to an ion on a mineral surface.
[20]
26
|
Colorado School of Mines
|
2.3.2 Modifiers and Depressants
Collectors alone are rarely sufficient to achieve desired selectivity in flotation. A
variety of compounds exist to modify the surface and/or solution chemistry of a mineral
flotation system. The pH must be regulated to create the proper electrokinetic
environment for the collectors to attach to the minerals. As bastnaesite flotation occurs
in alkaline environments, a base (hydroxides and soda ash are common) must be
added to raise the pH. Other chemicals are used to inhibit flotation of a specific mineral;
these are depressants.
2.3.2.1 Soda Ash
Soda ash (Na CO ) is used to regulate pH. The carbonate ion is a potential-
2 3
determining ion for bastnaesite and calcite. In the flotation window of 8 < pH < 10, HCO
-
- and CO 2- are present in solution. At high enough concentrations, the surface of barite
3 3
is converted to a barium carbonate surface. Excess soda ash can lead to depression of
bastnaesite. [22]
2.3.2.2 Lignin Sulfonate
Lignin sulfonate is a complex compound derived from the sulfonation of lignin, a
cellulose binder, in wood pulp. It is known to flotation as a barite depressant. [21], [22],
[26] Its selectivity for barite at high pH is likely due to the highly positive surface of the
barite compared to that of bastnaesite and calcite. It has also been proposed that the
molecule fits better on the barium sulfate structure as compared to the carbonate
structures of bastnaesite and calcite.
Bastnaesite recovery as a function of ammonium lignin sulfonate concentration is
given in Figure 2.23. Barite flotation with 1x10-5 kmol/m3 sodium oleate is virtually
eliminated by 1x10-5 kmol/m3 ammonium lignin sulfonate.
2.3.2.3 Metal Salts
Inorganic salts can increase flotation by binding and providing new adsorption
sites or decrease it by competing with collector ions. Salts composed of potential-
determining ions can manipulate the charge of minerals in the system, such as SO 2- for
4
29
|
Colorado School of Mines
|
2.4 Gravity Concentration
Often before flotation is considered, gravity separation is investigated as a
means of concentrating an ore. In the case of liberated, similarly-sized beach sands or
free gold deposits where the desired mineral is much denser than the undesired
minerals, gravity can quickly sort the valuable material from the gangue. This approach
relies on complete liberation – distinct particles of the valuable component must exist
without contact with gangue particles. These separations generally show better results
with a near-size deslimed feed. When a disparity exists between the specific gravities of
the two components, gravity may be used to sort them.
The Concentration Criterion is used as a first estimate of the relative success of
gravity separation. It compares the specific gravities of the heavy particles D , light
h
particles D, and fluid (usually water) D.
l f
Concentration criteria greater than 2.5 indicate that gravity separation is viable,
while those below 1.25 mean it is practically impossible. Values between those suggest
that using the right equipment and a carefully controlled feed, a separation could be
made. [39] Table 2.4 shows the concentration criterion for major components of the
bastnaesite ore used in this study.
Not every type of gravity equipment is suitable for a given application, which is
why so many exist. There are vibratory motion-based devices, such as jigs and shaking
tables, centrifugal units like Falcon and Knelson concentrators, and other devices:
spirals, multi-gravity separators, and heavy media separators. Characteristics of the
equipment such as allowable feed size (Figure 2.28), throughput, water requirement,
plant footprint, and power requirement dictate their applicability to a given separation.
Following the Concentration Criterion calculation, a float sink analysis may be done to
determine the efficiency of the separation as a function of specific gravity.
33
|
Colorado School of Mines
|
Table 2.4. Specific Gravity (SG) and Concentration Criterion (CC) for major components
of Molycorp ore.
Mineral Formula SG CC
Monazite (Ce,La)PO 5.2 2.47
4
Bastnaesite (Ce,La)FCO 5.0 2.35
3
Barite BaSO 4.5 2.06
4
Parisite Ca(Ce,La) (CO ) F 4.4 2.00
2 3 3 2
Rutile/Anatase TiO 4.1 1.81
2
Strontianite SrCO 3.8 1.65
3
Ankerite Ca(Fe,Mg)(CO ) 3.0 1.18
3 2
Chlorite (Mg,Al,Fe) [(Si,Al) O ](OH) 2.9 1.12
12 8 20 18
Calcite CaCO 2.7 1.00
3
Quartz SiO 2.7 1.00
2
Feldspar (K,Na,Ca…) X(Al,Si) 3O
8
2.7 1.00
These tools, along with preliminary testwork, can be used to assess the type of
equipment suitable to beneficiate an ore. In the case of rare earth ores, the recent
development of centrifugal-type concentrators has pushed the boundary of allowable
separation into the domain at which these minerals concentrate. As the minerals require
liberation, they must be ground finely. Excessive fines can be problematic in flotation
due to entrainment and agglomeration issues, although dispersants can be used to
mediate that effect. The slimes can also reduce the sharpness of the separation, as
seen in Figure 34. While potentially a problem for some types of equipment, spinning-
bowl concentrators are able to process feeds as fine as tens of microns. They are joined
by hydrocyclones, tilting frames, Mozley tables, and froth flotation as the only non-
magnetic unit operations, according to Figure 2.28, capable of handling a feed of that
size. Wet tables reach into the top end of this range, along with many other operations.
Two carbonatite operations similar to the Mountain Pass deposit that use physical
beneficiation methods are Sichuan Mianning (shaking tables and flotation) [10] and
Bear Lodge (scrubbing and sizing) [14].
Molycorp bastnaesite ore has seen prior attempts at shaking table concentration
in a lab setting. [40] The bastnaesite ore was pulverized and split into four fractions: 20-
38 μm, 38-53 μm, 53-75 μm, and 75-106 μm then purified with a Frantz Isodynamic
34
|
Colorado School of Mines
|
2.4.1 Heavy Liquid Separation
To quote Chris Mills, “The first step at the laboratory level should always be
heavy liquid analysis of the ore to be fed to the gravity separation plant.” The
information gained from such tests can dictate whether gravity separation will be easy
or difficult and which types of equipment are available to make the separation. [41]
While many fluids exist in the specific gravity range of 1.2 - 2.0, options available for
gravity separation of ores are both more limited and more expensive. [42] Historically,
potentially hazardous halogenated hydrocarbons have been used for this type of work.
Sodium polytungstate, a newer, nontoxic reagent with a maximum s.g. of 3.1
(adjustable by means of the water-to-powder ratio) was used in float/sink analysis. [43]
Figure 2.31. Density of aqueous sodium polytungstate solution as a function of mass
percent. [43]
2.4.2 Shaking Tables
Shaking tables are rectangular-shaped tables with riffled decks across which a
film of water flows. (Figure 2.32 and Figure 2.34) The mechanical drive imparts motion
along the long axis of the table, perpendicular to the flow of the water. [44] The water
carries the particles of the feed in slurry across the riffles in a fluid film. This causes the
fine, high density particles to fall into beds behind the riffles as the coarse, low-density
particles are carried in the quickly-moving film. (Figure 2.33) The action of the table is
such that particles move with the bed towards the discharge end until the end of the
37
|
Colorado School of Mines
|
The feed characteristics, feed rate, riffle pattern, and motion of the table should
all be tailored to fit the desired application. Feed characteristics are generally set by the
comminution circuit, but classifying, either with screens or cyclones, can influence the
separation on the table. Riffle pattern is most easily controlled by changing decks: often
a sands deck is used for coarse feeds and a slimes deck is used for fine feeds. The
drive of the table can be manipulated in both stroke length and frequency. A longer
stroke will require more water but moves heavies to the concentrate end more quickly.
Deister decks are set up at an incline from the drive to discharge end to allow migration
of heavies to the concentrate end and allow light particles to fall to the tails easier. The
tilt from the dressing side to the tailings side is generally maintained to allow a wide
spread of material at the concentrate end. While the tails-middlings cut point is dictated
by collection bins around the table, the concentrate-middlings cut point is made by the
operator at some point along the discharge end.
Figure 2.34. Wilfley shaking table
2.4.3 Knelson and Falcon Concentrators
Development of centrifugal concentrators was pioneered by those searching to
separate free gravity recoverable gold (GRG). In the 1970s, the Knelson Bowl (Figure
2.35) and later the Falcon Concentrator (Figure 2.37) were developed to use centrifugal
force to amplify the force of gravity for the purpose of separating constituents of an ore.
39
|
Colorado School of Mines
|
CHAPTER 3: EXPERIMENTAL METHODS
Once the literature survey was completed, an experimental campaign was
developed and carried out to determine the effectiveness of gravity separation before
froth flotation.
3.1 Characterization and Mineralogy Procedures
Ore was provided in two lots by Molycorp Inc. The first was crushed in a jaw and
roll crusher, then ground batch-wise in a laboratory ball mill until it was 100% passing
100 mesh. This ore was then blended and split in a Jones Riffle, and was used for the
microflotation tests. The second was crushed in a roll crusher until it passed 12 mesh
then was blended and split into representative samples in a Jones Riffle. Those
samples were wet ground in a rod mill for the required length of time.
Samples of the ore were sent to the Center for Advanced Mineral and
Metallurgical Processing (CAMP) at Montana Tech and to the Colorado School of Mines
Geology Department for MLA and QEMSCAN analysis, respectively. Several samples,
ground in a rod mill for specific lengths of time (zero, 10, 30, 60, and 90 minutes), were
characterized by CAMP to determine the liberation behavior of the ore components as a
function of grinding time. Elemental composition was determined by the automated
mineralogy software (as part of MLA and QEMSCAN), x-ray fluorescence, and x-ray
diffraction. The microflotation samples were analyzed with the Kroll Institute XRF, while
all of the gravity and magnetic test samples were analyzed by Hazen Research, Inc.
Size analysis was performed with a Microtrac Particle Size Analyzer and several Tyler
sieves.
3.2 Microflotation
Sodium oleate, octanohydroxamic acid, soda ash, ammonium lignin sulfonate,
sodium silicate, and copper (II) nitrate hemipentahydrate were procured as solids and
dissolved in de-ionized water to create stock reagent solutions. Concentrations (listed in
43
|
Colorado School of Mines
|
One-gram samples were taken in batches from a bag (approximately 15-20
grams per batch) and the grade of those batches was recorded. Conditioning was
performed in two separate 150-ml beakers. This was done to limit the amount of slurry
entering the collecting arm of the tube, therefore artificially inflating the recovery
numbers. The volumes were selected based on the volume of the cell. At around 40 ml,
solution began to pour into the collecting arm. As such, 25 ml of water was targeted for
the slurry beaker (ore, water, and depressant), although this number varied slightly as
more or less stock depressant solution was used to achieve the desired concentration.
The solutions were heated on a hot plate as the slurry was stirred with a stir bar. After
ten minutes (at a temperature of 80±10°C), the collector was added to the slurry and pH
was adjusted to 9.0±0.1 using drops of soda ash solution to follow plant practice. After
fifteen minutes, the slurry was removed from the hot plate and poured into the
Hallimond tube. The remaining solution was then poured in.
Flotation lasted for two minutes at a flow rate of 60 cc/min nitrogen gas. The
slurries were stirred with a stir bar to ensure mixing. The apparatus used for the tests is
shown in Figure 3.1. After that time, the concentrate was gathered by removing the
stopper from the collecting arm and flushing the froth from the tube. The remaining tails
solution was collected as well. Both the concentrate and tails were filtered through
Whatman 40 (8-μm pore size) filter paper and dried for 18 hours.
Figure 3.1. Modified Hallimond Tube used for microflotation tests.
45
|
Colorado School of Mines
|
3.3 Magnetic Separation
Fifty grams of ore ground for zero, 30, and 90 minutes were charged at 10%
solids through the WHIMS at different field strengths, based on percentages of
maximum amperage. The two-factor DOE matrix (designed with Stat Ease) is given in
Table 3.2.
Table 3.2. WHIMS DOE Parameters
DOE Field Strength,
Standard WHIMS Run P80, µm Gauss
5 1 762 5000
4 2 50 5000
6 3 762 10000
1 4 144 7500
3 5 50 10000
2 6 144 7500
3.4 Gravity Concentration
Sodium polytungstate heavy liquid medium was used to investigate the possibility
of beneficiation by gravity separation. Five grams of ore (ground for zero, 30, and 90
minutes) were centrifuged in 10 mL of fluid of specific gravity ranging from 2.70 to 2.95.
The floats were poured and skimmed off. The middling solution was poured off, and the
fines were flushed. Each product (floats, middlings, and sinks) was washed and filtered
several times and dried. The composition of the products was determined with x-ray
fluorescence.
A Deister table was used to develop a qualitative response to gravity
concentration. The process variables were adjusted based on visual observation. An
initial 500 g feed was tabled, yielding a concentrate, middling, and tailing product. That
concentrate was tabled again to represent a cleaning step. Two more 500 g batches
were tabled under the same conditions. From one of those tests, the concentrate,
middlings, and tailings were tabled again. A 300 g charge of 100 x 325 mesh ore was
also processed to investigate the effect of a classified feed. Fluidization water was set at
46
|
Colorado School of Mines
|
the minimum level that created a film across the entire table. The ore was slurried and
hand-fed from a small bucket into the feed box. Table tilt was adjusted as the feed
spread across the table. The cut point for concentrate and middlings was made at the
discretion of the operator based on the visual quality of the film. The table elevation,
stroke length, and stroke frequency were kept constant throughout all tests.
Stat Ease Design-Expert 9 was used to generate a Design of Experiments matrix
for work on the Falcon Concentrator. The factors chosen (Table 3.3) were G-Force
(controlled by the frequency on the Variac controller attached to the Falcon motor),
Feed Rate (controlled by opening the valve between one-half and one full turn open),
and Feed P in microns (controlled by grinding for specific lengths of time). The slurry
80
density was maintained at 10% solids for all tests, although that was approximated in
later tests due to uncertain moisture content. The slurry was mixed in a 20-liter tank and
fed through a valve to the Falcon.
Table 3.3. Falcon DOE Parameters
DOE Falcon G-Force, Grind Time, Feed Rate,
Standard Run G's minutes kg/hr
Opt 1 100 90 30
Opt 2 100 90 30
Opt 3 100 90 30
1 1 100 90 30
6 2 250 30 30
4 3 250 90 60
3 4 100 90 60
9 5 175 60 45
8 6 250 30 60
11 7 175 60 45
2 8 250 90 30
5 9 100 30 30
7 10 100 30 60
10 11 175 60 45
47
|
Colorado School of Mines
|
3.5 Bench Flotation
Bench Flotation tests were performed to compare the result of gravity
preconcentration to a circuit with no such step. A benchmark test using ore ground to a
P of 50 μm was carried out using lignin sulfonate and hydroxamic acid. Each test was
80
performed according to the timetable of Table 3.4. The slurry was heated at 33% solids
to approximately 85°C, and transferred to the flotation cell for conditioning. Ammonium
lignin sulfonate and octanohydroxamic acid were added and the motor turned on once
the slurry was in the heated cell. After fourteen minutes one drop of methyl isobutyl
carbinol was added to encourage frothing. At fifteen minutes, the air valve was opened
and flotation began. Flotation continued for ten minutes, with scraping of the froth
occurring every thirty seconds. For all tests, the air flowrate was 280 ml/min and the
motor stirred at 900 RPM. Following the tests, samples were filtered, dried, weighed,
and analyzed by XRF. Those conditions were repeated for a flotation test of the gravity
concentrate.
Table 3.4. Bench Flotation Timetable
Time Event
Start Beginning of conditioning: turn motor on;
add slurry, collector, and depressant; adjust
pH
14 minutes Addition of frother, adjust pH
15 minutes Beginning of flotation: Turn air on
16 minutes Scrape froth, (repeat every 30 seconds)
25 minutes End of flotation: turn air and motor off
48
|
Colorado School of Mines
|
CHAPTER 4: RESULTS AND DISCUSSION
The data received from the experiments was collected and analyzed, and is
presented below along with discussion of its interpretation.
4.1 Characterization and Mineralogy
Mineralogy results were received for the two lots of ore submitted for
characterization by MLA and QEMSCAN. The initial lot was used for the microflotation
and shaking table concentration and the second was used for the Falcon concentration,
magnetic separation, and comparative flotation tests. The results for the second lot are
presented in this section and those for the second lot can be found in FIRST LOT
MINERALOG.
The particle size distribution and P for each grind time is are given in Figure 4.1
80
and Table 4.1. These distributions were generated by grinding batches of ore for the
specific time (zero, 10, 30, 60, and 90 minutes). Subsequent test batches were ground
for similar lengths of time based on the intended distribution and size at which point
80% of the material passes (P ).
80
Figure 4.1. Particle size distributions for ground ore samples
49
|
Colorado School of Mines
|
Table 4.1. P80 of Ground Ore Samples
Grind Time (minutes) P (microns)
80
0 762
10 245
20 137
30 144
60 77
90 50
The composition was determined by MLA analysis and XRF. The modal mineral
content is shown in Table 4.2. Barite (24.6%), calcite (21.3%), dolomite (11.6%), and
quartz (7.65%) were identified as the dominant gangue minerals, with bastnaesite
(8.9%) and other rare earth minerals parisite (1.89%), monazite (0.99%), and allanite
(0.28%), making up the valuable content of the ore. Once the software calculated the
mineral content, it converted those numbers into elemental contents, which are
compared with the semi-quantitative XRF values in Table 4.3.
The barium, cerium, and lanthanum values according to MLA are between two
and four times as high as according to the XRF. The MLA sulfur concentration is slightly
higher than that given by XRF. The calcium, iron, and silicon numbers agree generally
well. The discrepancy between the methods, as well as the rise in REE content
associated with grind time, is assumed to be due to an overestimation of the dense
minerals, even though special care was taken to avoid such a bias in sample prep.
A false-color image of the largest size fraction of the unground ore is displayed in
Figure 4.2, and one of the 90-minute ground sample 200 x 400 mesh size fraction is
shown in Figure 4.3. The obvious particle size reduction, as well as the increase in
liberation due to grinding, is evident upon comparison of the two images.
The bastnaesite does not strongly report to a specific size fraction, as can be
seen from the mass distributions in Table 4.4. The bastnaesite grade tends to increase
slightly with decreasing particle size and with grinding time. This is an unusual
phenomenon, as it would be expected to remain constant because all samples were
taken from the whole in a similar fashion – the only difference was the grinding time.
50
|
Colorado School of Mines
|
liberation of the samples, upgrading to 71% and 69% of the minerals being 100%
liberated, and almost all of the mass (98%) being more than 25% liberated.
Table 4.4. Mass and Bastnaesite Distributions. The mass percentage for each size
fraction is given in plain text and the bastnaesite distribution in bold.
762 µm 245 µm 144 µm 77 µm 50 µm
60 56.9 -- -- -- -- -- -- -- --
+50 Mesh
50 X 100
14.4 13.5 34.7 30.5 8.1 4.2 0.2 0.1 0.1 0
Mesh
100 X 200
8.6 8.8 27.3 26.6 35.2 29.6 9.3 3.4 1.4 0.2
Mesh
200 X 400
5.7 6.4 13.1 12.7 20.3 19.5 33 31.4 23.1 15.8
Mesh
-400 Mesh 11.3 14.3 24.8 30.2 36.5 46.7 57.4 65.2 75.4 83.9
Total 100 100 100 100 100 100 100 100 100 100
Table 4.5. REE mineral and calcite liberation as cumulative mass recovery. Bolded
values represent the combined total of REE minerals (bastnaesite, parisite, monazite,
allanite).
Percent Crushed Ore
Liberated 762 µm 245 µm 144 µm 77 µm 50 µm
100 22 16 43 39 60 57 65 59 71 69
75 45 65 73 79 81 85 86 88 89 90
50 65 79 84 90 90 93 92 94 94 95
25 85 92 94 96 96 97 97 98 98 98
The zeta potential of the ore in distilled water was determined using a Stabino
zeta potential device. The IEP of the ore was found to be 8.0 ± 0.3. IEP’s for barite,
calcite, and bastnaesite were found to be 6.0 ± 0.2, 6.2 ± 0.4, and 6.0 ± 0.3. [50]
53
|
Colorado School of Mines
|
The difference in IEP of the ore (Figure 4.5) from those of the minerals
composing it suggests that while the main components are barite, calcite, and
bastnaesite, the electrokinetic behavior of the ore is not similar to the behavior of any
one major component.
4.2 Microflotation
Microflotation data was plotted in terms of elemental recovery and grade vs
depressant concentration (or collector concentration when no depressant was used).
Collector concentrations were held at either 5x10-5 M sodium oleate or 3x10-4 M
octanohydroxamic acid for the depressant tests after a satisfactory response was
obtained with those middle concentrations. Different oleate and hydroxamate
concentrations were used after the literature survey showed that their effects were not
similar at equal concentrations Oleate data is presented as filled points and
hydroxamate data is shown with outlines. The experimental data for the flotation tests,
including mass balances and elemental accountabilities, is given in APPENDIX C:
EXPERIMENTAL DATA.
Reproducibility can be seen for the oleate system in comparing tests 2 and 13
and for the hydroxamate system in comparing tests H6 and H6b. These tests were
performed using similar concentrations at carried out at different times to ensure that
similar test conditions yielded repeatable results. However, the information from them
should be considered qualitative at best.
Figure 4.6 and Figure 4.7 illustrate the difference between the collectors in terms
of recovery and concentrate grade, respectively. Both show increases in recovery with
concentration, but little difference in grade. The increase in recovery with concentration
is expected: with more collector molecules in solution, more mass can be attached to
bubbles to float.
55
|
Colorado School of Mines
|
The REE and calcium recoveries in both systems fell, but the hydroxamate
system showed a stronger effect with respect to sodium silicate addition (Figure 4.10).
Again, Figure 4.11 shows no appreciable change in grade.
The depression of hydroxamate flotation of all minerals by sodium silicate is
attributed to the high pH reached during conditioning, as recovery is shown (Figure
2.22) to have a peak and sharp cliff at pH 10 and higher. The addition of sodium silicate
raised the conditioning pH of these tests to 10.0, 10.6, and 10.8 (oleate); and 9.6, 10.1,
and 10.2 (hydroxamate). Reduction of pH was not considered due to an effort of limiting
reagent use and high acid consumption by reaction with calcium.
The cause for the continuous drop in hydroxamate recovery is likely due to a lack
of collector adsorption, caused by the increase in pH by the sodium silicate. As the
mechanism for adsorption relies on the reaction of collector molecules and charged
hydroxy species, those species must be prevalent for flotation. But considering the
dominance of uncharged Ce(OH) at pH of 10 and higher (shown previously in Figure
3
2.3), the adsorption process breaks down. The result is reduced recovery. Reduction of
pH prior to conditioning was not used in an effort to keep experimental conditions simple
for these qualitative tests.
Figure 4.10. Elemental recovery as a function of sodium silicate concentration.
58
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.