University
stringclasses 19
values | Text
stringlengths 458
20.7k
|
|---|---|
Colorado School of Mines
|
120
50 degree C
100
room temperature
)
2
m
/ lo 80
m
µ
(
y
t
is 60
n
e
D
n
o
it
p 40
r
o
s
d
A
20
0
4 5 6 7 8 pH 9 10 11 12 13
Figure 6.38 Adsorption density of the sample at room temperature and 50°C in the
presence of 5Χ10-4 M octanohydroxamic acid
Table 6.5 Results for WHIMS rougher with steel balls
Current Iron assay (%) Iron recovery (%) REO loss (%)
4 26.5 55.7 14.9
8 21.0 70.7 36.5
12 20.4 71.3 39.0
16 20.4 72.3 39.7
6.4 Bench Scale Flotation
The results of a study to determine optimum flotation conditions for the sample
from Bear Lodge are described in this section. To establish a reagent scheme batch tests
were performed, and a comparison with rare earth grade and recovery was applied as
the assessment. Different dosages of hydroxamic acid and modifiers were tested.
Meanwhile, varied pH, impeller speeds, and size fractions of the ore were also tested.
6.4.1 Rougher Flotation Tests
After wet high intensity magnetic separation, a series of flotation tests as a function
of collector concentration were conducted at pH 9 when HCl and Na CO are pH
2 3
80
|
Colorado School of Mines
|
adjustment solutions. The results (Fig. 6.39) clearly shows that a decreasing trend of the
grade of REO takes place as the collector concentration increases. Whereas, the
recovery of REO makes an increase until the concentration of hydroxamic acid is at 5Χ10-
4 M, followed by a drop. It could be due to a fact that the more usage of hydroxamic acid,
the less selectivity obtained. Moreover, when the addition of octanohydroxamic acid is
above 5Χ10-4 M, the formation of micelle could happen, which contributes to the decrease
of recovery.
To investigate the effect of pH, rougher tests were carried out with modifications
to this variable using HCl and Na 2CO
3
in the presence of 5Χ10-4 M hydroxamic acid. Fig.
6.40 gives the effect of pH on REO recovery and grade. It shows a slight benefit for grade
in favor of maintaining the pH below 7.5, which could also be expected to assist in pure
ancylite microflotation; however, the REO recovery decreases sharply. Therefore, taking
both grade and recovery into consideration, pH 9 is optimum. Another experiment was
also conducted using HCl and KOH solutions as pH adjustment in the presence of 5Χ10-
4 M octanohydroxamic acid. The results obtained (Fig. 6.41) indicate that above pH 9.5,
the trend in the presence KOH behaves similar with the trend of Na2CO , the usage of
3
KOH, however, is way less than that of Na CO , thus, KOH and HCl were employed as
2 3
pH adjustment solutions And the results, shown in Table 6.6, obtained from the tests as
a function of the collector concentration at pH 9 indicate that 5Χ10-4 M hydroxamic acid is
the optimum concentration for the sake of grade and recovery.
Table 6.6 The effect of various collector concentration on REO assay and recovery in
the presence of HCl and KOH as the pH adjustment solutions
Concentration mol/L REO assay (%) REO recovery (%)
5Χ10-4 8.8 78.8
1Χ10-3 7.0 78.9
Since strontianite and calcite, the main gangue minerals associated with ancylite,
are carbonate minerals, the only difference between those three minerals is that the
cation ions, such as Sr2+ and Ca2+, are characteristic, which is also confirmed by zeta
potential tests. Therefore, a series of batch tests were performed as a function of the
addition of strontium ion. Strontium nitrate was added with the ore and then conditioned
81
|
Colorado School of Mines
|
with the collector solution for 15 minutes. Thus, the influence of dosage is shown in Fig.
6.42. REO grade and recovery vary directly with the addition of strontium nitrate. The use
of strontium nitrate slightly increases REO grade but at the marked expense of recovery.
Thus, 7.5Χ10-4M is taken as the optimum addition for the rougher flotation. Another test
was conducted to show the influence whether the 10 minutes preconditioning with
strontium nitrate and the sample was necessary before the addition of the collector. The
results in Table 6.7 show that preconditioning time has a negative effect on the grade and
recovery.
Table 6.7 The effect of strontium nitrate preconditioning time on REO assay and
recovery
Preconditioning time Assay (%) Recovery (%)
(minutes)
0 9.7 79.5
10 9.5 72.9
20 100
18 90
16 80
14 70
12 60 )
%
) %
( y 10 50
(
y r e
a v
s s A o c e
8 40 R
6 30
REO assay
4 20
REO recovery
2 10
0 0
0 0.0005 0.001 0.0015 0.002 0.0025
concentration
Figure 6.39 Effect of the collector concentration on REO assay and recovery in the
presence of Na CO
2 3
82
|
Colorado School of Mines
|
20 100
18 90
16 80
14 70
12 60
)
%
) %
( y 10 50
(
y r e
a v
s s A o c e
8 40 R
6 30
4 REO assay 20
2 REO recovery 10
0 0
0 0.001 0.002 0.003 0.004 0.005 0.006
Concentration (mol/L)
Figure 6.42 Effect of strontium nitrate on REO assay and recovery in the presence of
HCl and KOH as pH adjustment solutions
Moreover, to improve the flotation selectivity, more efforts were made on
investigating other depressants. Sodium fluorosilicate, commonly used in the rare earth
flotation industry, was also tested in varied concentration. Sodium fluorosilicate
depresses all the carbonate minerals no matter how much the concentration is. It is
probably due to the fact that the surface chemistry qualities of ancylite, calcite and
strontianite are too similar to distinguish any one from the others.
In an attempt to investigate the effect of the impeller speed on the flotation
performance, one test was conducted at 1300 rpm for conditioning and 1100 rpm for
floating. In comparison with the results obtained at 900 rpm for both conditioning and
floating, shown in Table 6.8, the increasing speed gives a slightly positive effect on the
grade and a slight deduction for the REO recovery. The reason is that the increasing
speed makes the solids more suspended so that there is a greater possibility that the
collector can adsorb on the mineral surface.
84
|
Colorado School of Mines
|
Thus, when the flotation condition is: 20% solid density, 5Χ10-4 M hydroxamic acid,
7.5Χ10-4 M strontium nitrate, pH 9 adjusted by KOH and HCl solutions, a 1300 rpm
conditioning impeller speed and an 1100 pm floating impeller speed, 10.7% of REO assay
is obtained with 78.3% of recovery.
On the other hand, a desliming test was conducted to show whether desliming
could benefit the flotation performance. The desliming test was investigated using a 400
mesh standard Tyler sieve. The feed with 400 grams sample was poured in a shaker
covered by a 400 mesh standard Tyler sieve and rinsed with tap water. The sample with
both minus 400 mesh and plus 400 mesh were filtrated and dried, separatively.
Approximately 60.5% of the feed is in the size fraction from minus 100 mesh to plus 400
mesh, which is regarded as the flotation feed to run rougher flotation under the same
optimum condition. Table 6.9 indicates that after desliming, both assay and recovery are
depressed, which could be attributed to two assumptions. The first assumption is that 400
mesh is not fine enough to be treated as the desliming level. The second assumption is
that the size range from 100 mesh to 400 mesh is too big to get the sample well-liberated.
Table 6.8 The effect of impeller speed on REO assay and recovery
Condition REO assay (%) REO recovery (%)
900rpm for conditioning
9.7 79.5
and floating
1300rpm for conditioning
10.7 78.3
and 1100rpm for floating
Table 6.9 The effect of desliming on REO assay and recovery
Condition REO assay (%) REO recovery (%)
No desliming 10.7 78.3
After desliming 7.0 56.9
6.4.2 Cleaner Flotation Tests
In order to further optimize the assay of REO, the cleaner flotation tests were
conducted as the functions of hydroxamic acid concentration and the usage of strontium
nitrate. Moreover, the regrinding process also was investigated to delineate the effect of
particle size for the cleaner flotation performance. The concentrate (10.7% REO) of 125
85
|
Colorado School of Mines
|
grams was employed using 1L Denver Cell with an 1100 rpm impeller speed for
conditioning and 900 rpm impeller speed for flotation. The air flow rate was 380 ccm. HCl
and KOH solutions were used to keep the pH constant at 9.
Two concentrations of hydroxamic acid were employed to show that the flotation
performance at 1Χ10-4 M hydroxamic acid was better than that at 2.5Χ10-4 M hydroxamic
acid, when taking both assay and recovery into account. Table 6.11 indicates that 5Χ10-4
M Sr(NO 3)
2
gives the higher assay and recovery, compared with that at 5Χ10-4 M Sr(NO 3) 2.
It is probably due to the depression of Sr(NO ) . Moreover, in order to get the larger
3 2
surface area, the regrinding process was conducted by grinding rougher concentrate so
that 100% of the concentrate passed 200 mesh. The results obtained, however, illustrate
that the REO assay is almost constant while the recovery decreases, compared with the
cleaner test without regrinding. It is attributed to the fact that overgrinding happens during
the regrinding process so that the particle is too small to be floated up.
Table 6.10 The effect of hydroxamic acid on the cleaner flotation
Collector Concentration REO Assay (%) REO Recovery (%)
(M)
2.5Χ10-4 10.7 78.6
1Χ10-4 10.9 76.8
Table 6.11 The effect of strontium nitrate on the cleaner flotation
Sr(NO ) usage (M) REO Assay (%) REO Recovery (%)
3 2
5Χ10-4 10.5 63.5
4Χ10-4 11.2 72.7
Table 6.12 The effect of regrinding on the cleaner flotation
Condition REO Assay (%) REO Recovery (%)
No regrinding 11.2 72.7
Regrinding 11.2 68.3
A conclusion could be made that a cleaner concentrate with 11.2% assay and 72.7%
recovery could be obtained under the condition of 1Χ10-4 M hydroxamic acid, 4Χ10-4 M
Sr(NO ) , 11.11% pump density, 0.0266g/kg AEROFroth 70 and an 1100 rpm impeller
3 2
speed for conditioning, as well as a 900 rpm impeller speed for floating.
86
|
Colorado School of Mines
|
6.4.3 Flotation Simulation
On the basis of the conclusion of the previous studies, locked-cycle tests were
supposed to be conducted. However, due to the limited amount of the Bear Lodge ore,
the locked-cycle tests cannot be experimentally conducted. Thus, a flotation simulation
was conducted based on the theory of JKSimfloat software. The flotation flowsheet
proposed was illustrated that the non-magnetic minerals after magnetic separation was
fed to the rougher flotation with a pump density of 20%, then the rougher concentrate was
sent to the cleaner stage with a pump density of 11.11%. The tailing from the cleaner
flotation was sent back to the rougher flotation. Thus, the cleaner concentrate and rougher
tailing were regarded as the final products.
In this simulation, the first order kinetic model, shown in Eq. 6.16, was used to
model the flotation behavior.
-dC/dt = kC (Eq. 6.16)
Where C = the concentration of the specific mineral
t = flotation time
k = first order rate constant
After integrating Eq. 6.16, the recovery in laboratory batch tests at each time could
be determined for minerals and the flotation rate was calculated using the following
equation.
R = 1-EXP(-kt) (Eq. 6.17)
Where R = overall mineral recovery
t = flotation time
k = first order rate constant
Alexander et al [83] indicated that the flotation rate was divided into several
floatability components such as ore floatability, bubble surface area flux, and froth
recovery factor. The flotation rate constant could be expressed as Eq. 6.18.
k = P S R (Eq. 6.18)
b f
Where k = overall flotation rate constant (min-1)
P = ore floatability
S = bubble surface area flux (min-1)
b
R = froth recovery factor
f
87
|
Colorado School of Mines
|
According to the JK floatability index, the definition of the ore floatability is the
probability of a particle corresponding to the froth phase of the flotation process [84]. An
assumption is used that the total amount of minerals with floatability P in the concentrate
i
and tailing of a flotation cell is equal to that in the feed [83].
In this study, the floatability component distribution of ancylite and gangue are
shown in Table 6.13 and Table 6.14, respectively. They were calculated by the
experimental results under the flotation time of 6 and 4 minutes for the rougher and
cleaner flotation tests, respectively. The floatability rate could be obtained by replacing
the recovery of the cleaner flotation and 4 minutes into the Eq. 6.17. Likewise, theoretical
recoveries of the ancylite and gangue for each flotation stage, shown in Table 6.15 and
Table 6.16, respectively, could be calculated. Six floatability components distributions of
both ancylite and gangue minerals at different flotation stages (appendix B) were
conducted, and the final assay and recovery, shown in Fig. 6.43, were achieved to be
constant where the REO assay was 12.0% and the REO recovery is 72.4%. Therefore,
the flowsheet proposed with the final assay and recovery is shown in Figure 6.44.
Table 6.13 Floatability components distribution of ancylite
Floating Non-floating
Floatability rate (min-1) 0.3246 0
Floatability component
91.31 8.69
distribution (%)
Table 6.14 Floatability components distribution of gangue
Floating Non-floating
Floatability rate (min-1) 0.2724 0
Floatability component
57.12 42.88
distribution (%)
Table 6.15 Calculated recovery of different stages of ancylite flotation
Calculated recovery (%)
Stage name Flotation time (min)
Floating Non-floating
Rougher 6 85.7 0
Cleaner 4 72.7 0
88
|
Colorado School of Mines
|
Table 6.16 Calculated recovery of different stages of gangue flotation
Calculated recovery (%)
Stage name Flotation time (min)
Floating Non-floating
Rougher 6 80.5 0
Cleaner 4 66.4 0
14 100
90
12
80
10
70
60
8 ) %
) %
( y 50
(
y r e
a v
s s A 6 o c e
40 R
30
4
REO Assay 20
2
REO Recovery 10
0 0
0 1 2 3 4 5 6 7 8
times
Figure. 6.43 The effect of closed-circle times on REO assay and recovery
6.5 Conclusion
Based on the previous studies, a brief process, shown in Fig. 6.45, is developed
to treat the sample from Bear Lodge. The brief process is that the run-of-mine ore with
around 4.5% REO is first fed to the comminution circuit. The product, which is 100% minus
100 mesh, undergoes wet high intensity magnetic separators with 8.75 amps/m2 current
flux. The magnetic concentrate contains around 3.4% REO. The conditioning treatment
for WHIMS tailings is carried out with 1.27 Ib/ton strontium nitrate, 0.64 Ib/ton hydroxamic
acid and 0.05 Ib/ton frother. The pH at this stage is around 9. The slurry containing 20%
89
|
Colorado School of Mines
|
CHAPTER 7
PRELIMINARY EVALUATION
The basic purpose of the preliminary evaluation is to assess the potential feasibility
of the flowsheet purposed from the economic perspective. In the attempt to make an
economic evaluation, the estimates of the capital and operating costs of the processing
plant were made. Capital cost estimates are generally divided into two portions: a fixed
capital and a working capital. The fixed capital in this study was estimated via the O’ Hara
method. Working capital cost estimation was determined following the rule that the
working capital cost is 12% to 15% of the fixed capital. The operating cost was estimated
by the comparative cost estimate methodology. The cost estimate is likely to be within
±30% accuracy. Finally, a financial analysis was made to provide a description of
discounted cash flow analysis.
7.1 Capital Cost Estimate
It is well-known that capital cost consists of fixed capital and working capital. In this
study, the fixed capital was estimated by the O’Hara method, and the working capital is
calculated by the method mentioned by Mular et al [85], which is that the working capital
is 12% to 15% of fixed capital. Thus, the working capital being equivalent to 12% of the
fixed capital was used. Several assumptions are made for the capital cost estimation.
1. The mineral plant is built in a flat site.
2. The foundation is built on the solid rock.
3. Twenty seven employees are required in this mill.
4. Employee lives in bunkhouses.
5. The capacity is 1,100 tons per day.
Other than several assumptions, the 2014 technical report of Bear Lodge Project
[59] is regarded as the reference to estimate the capital cost. The water pipe length of
1,500ft is based on distance between the wells and the PUG plant. The electricity was
assumed to be provided by PreCorp. Therefore, a capital estimate is listed in Table 7.1.
The total current capital cost is calculated by the reference cost multiplied by the
index ratio. Thus, the total current capital cost is $62,088,000.
92
|
Colorado School of Mines
|
Table 7.1 Summary of the mill plant capital cost estimation
Cost Item Factor Reference Cost (1000$)
Clear/ excav. 1 690
Foundation 1 1,374
Crush/conv. - 3,092
Mill bldg. 1.8 3,711
Grind/storage 1.5 3,283
Flotation/WHIMS 3 2,052
Thicken/filt. 1 344
Con. Storage - 935
Tail pond - 206
Power lines P=20000 3,724
Plant services 27 243
Townsite housing 27 1,174
Infrastructure Feasibility, plan, design 0.07 99
Supervision, camp 0.09 127
Admin, staff 0.055 78
Pipe costs L=0.28 110
Water Fresh water pumps - 322
Reclaim water pumps - 420
Feasibility - 1,858
Supervise/camp 0.09 1,824
Admin, staff 0.055 1,115
Working capital 0.12 3,213
Total capital cost - 29,994
7.2 Operating Cost Estimate
The comparative estimate method is employed to make an estimation for the
operating cost. The comparative model reference is obtained from CostMine 2013 [86].
Complementary assumptions are made and listed below:
1. The flowsheet in this study shows that the run-of-mine ore is sent to the crushing circuit.
A Jaw crusher produces a product of minus 4 inches in size. A 4 in Χ 8 in vibrating screen
with 0.75 inch opening returns the oversize to a 3 ft diameter secondary standard core
crusher, which has 0.25 inch discharge; meanwhile, the undersize is sent to storage bins.
The product from the crushing circuit is sent to the grinding-magnetic separation-flotation
(G-MS-F) plant. Grinding is carried out in a 10 ft Χ 18 ft ball mill and produces a minus
100-mesh product for magnetic separation. The closed grinding circuit also contains a 6
93
|
Colorado School of Mines
|
inch cyclone. The overflow undergoes magnetic separation. The underflow is sent back
to ball mill. The tailing and concentrate from magnetic separation are pumped to two 20
ft thickeners. The tailing undergoes conditioning and flotation treatments. The concentrate
from the cleaner flotation stage is transported to a thickener, filter disk, and rotary dry.
The final dry concentrate is stored in a front end loader. The tailing from cleaner stage is
pumped back to the rougher stage. The final tailings from the rougher flotation combined
with concentrate from the magnetic separation circuit are thickened and pumped to a
tailing pond.
2. The plant schedule is shown in Table 7.2.
Table 7.2 The plant schedule
Circuit Hours/day Days/week
Crushing 16 5
G-MS-F 24 7
3. The density of the run of mine is 3,500kg/m3.
4. Ball mill operates at 70% solid by weight.
5. Magnetic separation operates at 20% solid by weight.
6. Seventy-five percentage solid of the feed is in the overflow.
7. The tailing from magnetic separation contains 17.1% solid by weight, while the
concentrate contains 46.31% solid by weight.
8. The thickener underflow for magnetic separation tailings contains 60% solid by weight.
9. The rougher flotation operates at 20% pump density, while the cleaner stage operates
at 11% pump density.
10. The concentrate from the cleaner stage contains 80% solid by weight.
11. The total electricity contains the electricity used by main equipment and others. Others
is assumed to be 15% of the electricity used by main equipment.
12. The usage of fuel is assumed to be 300 gallon/day at the price of $2.78/gal.
13. The price of AERO 6493, which is a hydroxamic acid product developed by Cytec, is
$5.60/Ib. The price of MIBC and strontium nitrate are $1.5/Ib and $0.60/Ib, respectively.
14. Hourly personnel requirement and salaried personnel requirement, which is same as
a 1,000 tonnes model from CostMine (2013) [86], are listed in Table 7.3.
94
|
Colorado School of Mines
|
15. PreCorp provides a cost of electricity of $0.068 per kWh. [59]
16. Grinding media cost is assumed to be $750/ton.
The equipment and the supplies used in this plant are listed in Table 7.4 and Table
7.5, respectively.
Table 7.3 Hourly personnel requirement and salaried personnel requirement
Hourly Personnel requirement Salaried personnel requirement
Class Workers/day Mill superintendent 1
Crusher operators 2 General foreman 0
Grinding operators 2 Maintenance foreman 1
Magnetic separation 2 Plant foreman 3
operators
Flotation operators 3 Senior metallurgist 0
Filter operators 2 Metallurgist 1
Dryer operators 0 Process technician 1
Assayers 1 Instrument technician 1
Samplers 3 Process foreman 1
Laborers 6 Total salaried personnel 9
Mechanics 3
Electricians 3
Total Hourly Personnel 27
Table 7.6 shows the operation cost for the plant. Supplies are the greatest cost
followed by labor and administration.
7.3 Economic Analysis
Economic analysis for this study was undertaken utilizing the Discounted Cash
Flow (DCF) methodology and was based on the capital and operating cost estimates. The
capital and operating cost estimates are described in Chapter 7.1 and 7.2. The cash flow
model includes the following assumptions:
1. All amounts are constant dollars, not adjusted by inflation.
2. A constant price for the final concentrate is assumed to be $500/ton.
3. The assay and production of run-of-mine keeps constant.
4. There is no fluctuation for costs of labor, power and reagents. Tailing storage and
general administration costs keep constant over the 10 years.
5. Financial periods are equal to one year.
95
|
Colorado School of Mines
|
6. No capitalized equipment replacements are included after the beginning of production.
7. Severance tax rate and federal tax are 2% and 20%, respectively. The exemption for
federal tax is $40,000.
Table 7.4 List of the main equipment
Items Model HP Number
Jaw crusher 24 in Χ 36 in 125 1
Standard cone 3 ft diameter, ¼ ft
200 1
crusher discharge
4 ft Χ 8 ft, double
Vibrating screen 5 1
deck
Belt feeder 24 in Χ 4 in 3 2
Bin 8 ft Χ 8 ft 5 1
Ball mill 10 ft Χ 18 ft 1,000 1
Items Model HP Number
Cyclone 6 in diameter 1
High intensity
1 start, 3 passes,
induced roll 10 5
8 tph
magnetic separator
Thickener 20 ft diameter 1.5 4
500 gpm 15 5
200 gpm 10 1
Medium slurry pump 100 gpm 2.5 1
50 gpm 1.5 1
20 gpm 1 2
36 in impeller
Mixer 10 4
diameter
Maximum feed rate
Reagent feeder 0.08 8
0.26 gpm
Flotation cell 60 cu ft 10 24
30 in belt width, 60 ft
Belt conveyor 15 1
length
30 in belt width, 200
40 2
ft length
Filter disk 259 sq ft 1.5 1
4 ft diameter, 30 ft
Rotary dry 20 1
length
7.4 Sensitivity Analysis
There are numerous risks to the financial viability of this study. Sensitivity analysis
was performed to assess the impacts on the financial results. Several factors were
96
|
Colorado School of Mines
|
CHAPTER 8
SUMMARY AND CONCLUSIONS
The major objective of this dissertation is to separate the rare earth minerals from
the Bear Lodge. The first step in this endeavor was to complete the mineralogical
characterization of the ore in order to identify its composition, mineral association and
liberation. Mineral identification showed that ancylite was the main rare earth containing
mineral and associated with strontianite and calcite. Thus, an investigation for the surface
chemistry of ancylite, strontianite and calcite in the presence of hydroxamic acid, including
their surface charges, adsorption densities and micro-scale flotation behaviors, were
conducted. Particularly, the mechanisms for the uptake of hydroxamic acid on the three
minerals were examined. Furthermore, bench scale flotation tests were also conducted
to develop a flowsheet that can effectively extract rare earth from the sample of Bear
Lodge. Finally, a preliminary economic study was carried out to estimate the economic
value of the flowsheet proposed based on the lab-scale results. The following discussions
summarize the accomplishments and contributions.
Mineralogical characterization found that the major phase in the sample was
calcite with 60%. Minor minerals were pyrite (7.6%), ancylite (7.3%) and strontianite
(5.8%), with numerous trace species. Ancylite was the dominant rare earth mineral,
followed by bastnaesite and monazite. Ancylite was strongly associated with calcite and
strontianite, and the grain size of ancylite (P ) was around 50µm, while the ground
80
carbonatite had a P of 100µm. Because of the mineral specimen and associations,
80
fundamentals of surface properties for ancylite, strontianite and calcite were investigated
regarding the limited amount of literature on ancylite flotation.
The first approach of the fundamental studies was to investigate its surface
property to understand the flotation behavior of ancylite in the presence of hydroxamic
acid. The surface charge characteristics were conducted using a Stabino® distributed by
Microtrac Europe GmbH. Zeta potential results indicated that the isoelectric points of
ancylite, strontianite and calcite in distilled water were around 5.46, 4.5 and 5.5,
respectively. The effect of lattice ions of these three minerals were investigated and it is
found that Sr2+, CO 2- and HCO - were the determining ions for ancylite. The electrokinetic
3 3
100
|
Colorado School of Mines
|
results of the three minerals in the presence of hydroxamic acid showed that
chemisorption happened when hydroxamic acid adsorbed on the three minerals, which
was in accordance with the results from both FTIR and adsorption density measurements.
As the concentration of hydroxamic acid increased, isoelectric points of three minerals
decreased, compared with those in distilled water. Moreover, a scenario was observed
that the dissolved species affected the flotation behaviors.
The uptake of hydroxamic acid on ancylite, strontianite and calcite has been
determined as a function of time, concentration and pH, as well as temperature. The
results indicated a chemisorption happening on the mineral surface with hydroxamic acid.
Hydroxamic acid preferably adsorbed on the surface of ancylite when compared with
strontianite and calcite, which was due to the higher stability constant of rare earth
hydroxamate. The standard free energies of adsorption of hydroxamic acid were
calculated to be -6.15, -4.93 and -5.58 Kcal/mole for ancylite, calcite and strontianite,
respectively. The experimental and thermodynamic results at 50°C demonstrated that
strontianite and calcite were more sensitive with temperature, compared with ancylite.
The microflotation experiments showed that there was a difference between the
floatabilities of ancylite, strontianite and calcite as functions of hydroxamic acid
concentration and pH. However, due to the interference of dissolved species, the flotation
behavior of the mixture of these three minerals was different from the flotation behavior
of individual minerals.
The bench scale flotation experiments conducted in this research showed that it
was difficult to separate ancylite from calcite and strontianite, because of their similar
qualities of surface chemistry. However, the addition of strontium nitrate favorably
selectively separated ancylite from the other two minerals. Since hydroxamic acid
preferably adsorbs on iron surface and there is a high content of iron minerals, a wet high
intensity magnetic separation (WHIMS) was used before the flotation. After the ore
sample was subjected to wet high intensity magnetic separation (WHIMS), a non-
magnetic product was obtained which was fed to flotation. Flotation following a rougher-
cleaner process produced a concentrate containing 12.0% REO at a 61% recovery.
Based on the experimental results and assumptions, a preliminary economic study was
conducted as well. The assessment yielded an after-tax internal rate of return (IRR) of
101
|
Colorado School of Mines
|
ABSTRACT
Economic deposits containing ores readily amenable to processing, like free-
milling gold or gold occurring in oxides, are rapidly being depleted. This trend has
caused many gold mining companies to turn their attention to refractory ores. Processing
refractory gold ores requires the use of pretreatment processes that liberate gold from the
sulfide matrix or chemically alter the matrix in order to facilitate gold extraction. There
are several primary processes used by industry, including roasting, pressure oxidation,
and bio-oxidation. Each of these processes has specific advantages and disadvantages.
Bio-oxidation, for example, is often referred to as a “marginal and challenging” process
because it requires a long residence time and certain amount of expertise and skills from
the operator for successful implementation.
This dissertation aims to improve the economics and efficiency of the whole-ore
bio-oxidation technology by separating particles with favorable characteristics from
particles with unfavorable characteristics as part of a pretreatment process. To achieve
this objective, research was conducted to characterize all relevant properties of particles,
systematically distinguish favorable/unfavorable properties, and then design an
appropriate mineral processing method for each category. As a result, four properties
were identified as favorable for bio-oxidation. They included iron content, sulfur content,
specific surface area of pyrite, and gold content. In contrast, two unfavorable properties
for bio-oxidation are carbonate content and organic carbon content. Experimentation
revealed that particles favorable for bio-oxidation also had a higher density than particles
possessing unfavorable properties. This phenomenon allows for the utilization of gravity
separation technologies as a means of separating particles conducive to bio-oxidation
from those that are not. Experiments on the amenability of gravity separation in this
application were also performed. Experimental work resulted in the development of a
new design methodology that improves the overall efficiency of the whole-ore heap bio
oxidation process. Implementation of gravity separation technology as a pre-separation
tool decreases variations in bio-oxidation performance, increases the capacity of bio
oxidation pads by eliminating a portion of ore which can not be bio-oxidized very
efficiently, and increases the quantity of ore treated and gold recovery per unit area (pad).
|
Colorado School of Mines
|
ACKNOWLEDGMENTS
I would like to thank the many individuals who helped to make this dissertation
possible. First of all, I wish to express my sincere appreciation to the thesis advisors: Dr.
Rozgonyi who has led me through my research all these years and been more than just an
advisor to me; Dr. Miller for his strong encouragement as well as his advice and support.
A heartfelt thank you to Dr. Cohen, whose past experience and expertise on the subject of
my dissertation were invaluable. I would like to acknowledge Dr. McKinnon for
reviewing the mathematical model and adding valuable input; Dr. Kuchta for his
assistance in the preparation of the literature review for this manuscript; and
Dr. Khindanova for leading me through the milestones at the department of Economics
and Business. I am grateful to Dr. Olson and Dr. Clark for helping me to get started by
supplying microorganisms and consulting me about the practical aspects of bio-oxidation;
Dr. Brierley for taking time away from running the consulting firm to answer my
questions and providing me with all necessary technical support and information;
Prof. Thomas for reading my dissertation, contributing valuable input and providing me
with analytical procedures; Mr. Le Vier for providing me with a sample.
It would be unfair not to express special and warm thanks to my parents,
Mynzhassar and Saltanat, my brother and sister, Amir and Aigul, and my lovely wife
Sandugash and sons, Dimash and Nurmash, who have been supportive of my dreams all
these years.
xi
|
Colorado School of Mines
|
CHAPTER 1
INTRODUCTION
1.1 Problem Scope and Discussion
High grade economic deposits, including resources containing free-milling gold
or gold occurring in oxides, are rapidly being depleted. Consequently, it is becoming
increasingly common for gold mining companies to acquire and process refractory ores.
The following factors make gold ores refractory (1):
• submicron size of gold particles;
• tight association with sulfide matrix;
• carbonaceous matter in the ore;
• interference with other metals.
Processing refractory gold ores requires the use of pretreatment processes that
liberate gold from the sulfide matrix, i.e., chemically or mechanically alter the matrix in
order to facilitate gold extraction. There are several primary processes used by industry:
these include roasting, pressure oxidation, and bio-oxidation. Each of these processes has
specific advantages and disadvantages. Bio-oxidation, for example, is often referred to as
a “marginal and challenging” process because it is time-consuming and requires certain
expertise and skills from the operator for successful implementation. For this reason,
J. Marsden and I. House referred to biological oxidation methods as “proven
commercially for flotation concentrates; unproven for whole-ore treatment.” (1)
A process of interest that is currently being used in Nevada by a prominent
mining company is presented in Figure 1.1 (2-4). Run-of-Mine (ROM) ore is hauled by
trucks to a crushing facility, where it is reduced in size to Pgo=19 mm. The crushed ore
follows one of the two paths, as depicted in Figure 1.1. It may go directly to a flotation
mill, or the ore may undergo heap bio-oxidation as a pretreatment step before it goes to
the mill. Regardless of the path, the ore is milled and floated in order to separate sulfides
from non-sulfides. While the sulfide concentrates go to the roaster, tailings are subjected
to a carbon-in-leach process to recover freed gold by cyanide leaching. A detailed
1
|
Colorado School of Mines
|
This research aims to improve the economics and efficiency of sulfidic/refractory
gold ore processing by developing an innovative approach that increases the overall
performance of the whole-ore (run-of-mine) heap bio-oxidation. The ore for this research
was generously provided by one of the major mining companies operating in Nevada.
1.2 Dissertation Objective
The primary objective of this dissertation was to improve the process efficiency
associated with the heap bio-oxidation of refractory gold ores through the development of
a new design methodology. As envisioned, this research aims to advance the entire
processing cycle for these types of ores.
This objective was achieved by performing the following tasks:
• analysis of existing commercial operations;
• analysis of the state of the application of bio-technology to Mineral Processing;
• ore identification study;
• theoretical study;
• mathematical modeling;
• gravity separation amenability study;
• bio-oxidation amenability on various gravity fractions;
• mineral jigging amenability study.
1.3 Original Idea and Contribution to the Field of Study
The concept behind this dissertation is based upon on the premise that crushed ore
contains particles possessing distinctive physical, electrochemical, and mechanical
properties. The combination of these properties renders a very diverse mixture of
particles, where some respond to a particular treatment while others do not. This
difference in response leads to the conclusion that there are ore particles with favorable
and unfavorable characteristics for bio-oxidative pretreatment. If we can characterize all
relevant properties, distinguish favorable properties, and match them with the appropriate
mineral processing method, it will then result in a significant improvement of the entire
processing cycle.
3
|
Colorado School of Mines
|
Through this research, six primary ore properties influencing bio-oxidation
performance were identified. Any given ore particle might possess four favorable
properties: iron content, sulfur content, specific surface area of pyrite, and gold content.
In addition, any ore particle might also possess two unfavorable properties for bio
oxidation: carbonates content and organic carbon content. Fortunately, a direct
relationship appears to exist between these favorable properties and particle density. This
phenomenon enables the utilization of gravity separation technologies as a means of
separating particles conducive to bio-oxidation from those that are not. This theory has
been experimentally verified as a part of this research and represents a major
contribution.
Experiments on the amenability of gravity separation in this application were also
performed. As a result, fractions with different densities were obtained. Special tests and
analytical procedures were conducted on these fractions to obtain experimental data that
could be used to model and design flowsheets. Experimental work resulted in the
development of a new design methodology that improves the overall efficiency of the
whole-ore heap bio-oxidation process.
After a comprehensive literature review and consultations with industry experts, it
became quite apparent that the whole-ore heap bio-oxidation process requires substantial
technical improvements before it could be economically viable. The new approach,
combining gravity separation with heap bio-oxidation, as developed in this dissertation, is
intended to address these technical challenges. As the literature search shows, no
previous research on this subject is known to exist. This research yields theoretical
principals, experimental and analytical procedures, and a method of designing heap bio
oxidation circuits that are novel and have direct industrial application.
1.4 Reliability of Experimental Data and Conclusions
All conclusions drawn by the author are based on commonly accepted theoretical
principles and his own extensive experimental work. All discussions, recommendations,
and final conclusions are supported by experimental verification on different scales using
established industrial practices. The reported experimental values are processed
according to best industry standards and procedures. Samples used in this research
4
|
Colorado School of Mines
|
underwent special preparation procedures to ensure representative sampling. Glassware,
chemical grade reagents, and supplies used in this research were purchased from well-
known and reputable companies. All analytical equipment went through special
procedures to determine limitations and accuracy. Chapter 3 discusses the analytical and
experimental set-ups in detail. All analytical protocols were first tested on samples with
known composition to ensure applicability and accuracy. In addition, obtained results
were validated by sending samples to different certified commercial laboratories.
1.5 Organization of Dissertation
This dissertation comprises eight chapters. The first chapter of the dissertation
discusses general issues: the scope of the problems, the novelty of obtained results, the
outline of the dissertation, as well as steps taken to ensure the highest accuracy in
collecting the experimental data. The second chapter reviews available literature sources
and discusses the theory and practice of gold extraction from refractory ores. The third
chapter addresses the experiments’ design, procedures, and the set-ups involved in this
research. The fourth chapter describes the samples used, how they were obtained, and the
geochemical, mineralogical, and other analyses performed in order to accurately identify
the sample composition and properties. Chapter 5 uses the results obtained from
Chapter 4 to analyze and demonstrate the theoretical principles, such as thermodynamics
of a given aqua system and the kinetics of bio-oxidation. Chapters 6 and 7 present the
results of the research program on gravity separation and bio-oxidation tests respectively.
Each of these chapters contains a brief summary. The last chapter (Chapter 8) details the
overall conclusions of this work and recommendations for further research. The
remaining ancillary sections of this dissertation include cited references and appendices.
5
|
Colorado School of Mines
|
CHAPTER 2
LITERATURE REVIEW
2.1 Principles of Gold Extraction
Gold has favorable chemical (resistance to corrosion), physical (superior electrical
conductivity), and mechanical (malleability) properties that give it broad use in industrial
applications. Similarly, its relative scarcity and physical appearance have contributed to
gold’s wide spread use as a monetary instrument and adornment for centuries. The yearly
production of gold world-wide is approximately 2,800 short tons (st) (5). More than
140,000 st of gold have been produced in the world so far, and consumption is still
increasing (6). A majority of the gold produced is used in the following applications:
computers/semiconductors, spacecraft, telephones, TVs and VCRs, medicine/dentistry,
eye surgery, lasers, and jewelry. The jewelry-making industry consumes more than 80%
of the world’s gold production (5, 6). Gold is also widely consumed for investment
purposes, accounting for approximately 10% of the world’s consumption (5).
Figure 2.1 depicts a simplified, generic processing flowsheet for gold extraction
from different types of ores. Dash lines represent alternative flows of material, and solid
lines represent mandatory flows of material. As shown in Figure 2.1, the processing of
gold ores can be broken down into four distinctive stages: gold preconcentration (1), gold
extraction (2), crude bullion production (3), and high purity gold refining (4).
There are a number of gold preconcentration methods commonly used:
1. flotation and gravity concentration;
2. hydrometallurgical methods without oxidative pretreatment;
3. hydrometallurgical methods with oxidative pretreatment.
Some ores contain gold that cannot be easily recovered. Such refractory ores
require an oxidative pretreatment operation before or after a preconcentration/separation
process takes place. The following pretreatment methods are commonly used in the
mining industry (1):
1. biological oxidation;
2. pyrometallurgical oxidation;
6
|
Colorado School of Mines
|
3. hydrometallurgical oxidation.
Gold processing generally yields a solid product called a concentrate or a solution
enriched by gold, referred to as a pregnant solution. If the final product is a concentrate, it
goes to a smelter (the third stage in Figure 2.1), where crude bullion bars are produced
prior to refining.
Smelting with fluxes is carried out at an elevated temperature (7-9). A flux, which
facilitates slag-forming, is added to a gold-bearing material. Smelting normally takes
approximately 1.5 hours at a temperature range of 1,100-1,400°C. Once slag is
successfully formed, it is poured off, and then the precious metal alloy is recovered from
the furnace. Another type of pyrometallyrgical refining is called the “Tavener Process”
which can be characterized as a large version of the conventional fire assay procedure.
This process is used occasionally to refine low-grade products. Smelting is commonly
used to produce dore bullion bars at a mining location, while final purification takes place
in an off-site gold refinery.
Conversely, if the product of processing is a pregnant solution, it goes to a gold
extraction stage as shown in Figure 2.1. To recover gold, there are two alternatives used
by industry: adsorption on activated carbon or adsorption on ion-exchange resins. The
adsorption process is followed by gold stripping. Stripping desorbs gold and converts it
into a smaller volume of high-grade solution. This solution is then taken to the crude
bullion production stage to be treated by either electro winning or zinc precipitation.
Smelting produces crude bullion bars. Dore bullion bars are usually shipped directly to a
refinery (7-9).
A refinery produces high purity gold. To do that, a number of methods are
utilized. The chlorination process, sometimes referred to as the Miller process, was
developed by Dr. F. B. Miller in 1867 (7-9). This method is used to refine gold that does
not contain platinum group metals when a purity of 99.0-99.5% of gold is required.
Chlorine gas is injected through graphite or ceramic pipes into the molten metal at
1100°C. The gas forms the molten chlorides of silver and copper as well as the volatile
chlorides of other base metals. The volatile chlorides leave the gold as off-gas;
meanwhile, the chlorides of copper and silver float to the surface of the molten gold as
slag. The slag is skimmed off, and the endpoint of the reaction is indicated by the
|
Colorado School of Mines
|
appearance of reddish gold fumes. The appearance of reddish colored vapor indicates that
the molten metal contains more than 99% of the gold. Usually, it takes no more than a
few hours to complete the refining process.
Dr. E. Wohlwill invented an alternative process in 1878 (7-9). It requires using
raw gold as anode plates, and tetrachloroauric and hydrochloric acids as electrolyte. Gold
is then deposited on thin titanium cathodes; while copper, platinum, palladium, and other
constituents are dissolved in the electrolyte. Silver forms chloride and precipitates into
the anodic slime at the bottom of the tank. The process usually takes 6-8 hours and yields
gold of 99.99% purity. Further refining up to 99.999% can be done by using this process.
The sulfuric acid parting process, a hydrometallurgical refining method, was first
introduced in the early 1800s (7-9). It makes use of the fact that silver and other
impurities dissolve in sulfuric acid, whereas gold does not. The silver-gold ratio of the
bullion must be between 2:1 to 5:1. Granulation is preferred before the refining process
occurs. The refining takes 5-6 hours at the boiling temperature of sulfuric acid. The solid
residue is washed off and retreated with boiling acid several times. Ultimately, the
residue contains 99.6%-99.8% gold.
The Minataur process was first introduced in 1997 (7-9). Electro wining sludge
containing 50-85% gold, 8-10% silver, and 3%-6% copper was leached by chloride-
chlorine media to yield a gold-enriched solution containing 60-75 g/L gold. Solvent
extraction follows the leaching process to selectively recover gold from the aqueous
solution. After stripping, gold is precipitated by sulfur dioxide. This results in the
precipitant containing up to 99.99% gold.
2.2 Refractory Gold Ores Pretreatment Methods
There are three main pretreatment methods that have been developed to treat
refractory gold ores by the mining industry (10-21):
1. roasting of concentrates or ores;
2. hydrometallurgical oxidation of concentrates or ores;
3. bio-oxidation of concentrates or ores.
The oldest and most established pretreatment method among these is roasting.
9
|
Colorado School of Mines
|
Oxidative roasting possesses a short retention time (less than one hour) and a high
process efficiency (up to 99% of sulfide sulfur is oxidized). In addition, carbonaceous
matter is oxidized and removed from the material. Thus, roasting addresses the
phenomenon called pre-robbing. Moreover, this technology does not require water which
is often problematic in dry climates or where access is limited. On the other hand,
roasting requires the construction of sophisticated systems to clean the off-gas
contaminants. As a result, an oxygenated roasting facility requires a large capital
investment. An economic analysis comparing the operating cost of oxygenated roasting
versus an acid autoclaving facility was conducted by the company and showed that the
operating cost for roasting was superior to that of autoclaving (13).
Hydrometallurgical oxidation methods are based on the solubility of sulfide
minerals in an aqueous solution in the presence of an oxidant over a specific range of
temperature and pressure. There are four commercially proven oxidation technologies
(1):
1. low-pressure oxygen pre-aeration;
2. high-pressure oxygen - acidic media or non-acidic media;
3. nitric acid;
4. chlorine/chlorination.
The second category, pressure oxidation technique, can treat both concentrates
and ores and is broadly used in industry. Figure 2.3 represents an acid pressure oxidation
flowsheet utilized by Barrick Goldstrike to make sulfide minerals decompose. This
flowsheet is designated to treat the entire run-of-mine ore production (18).
The ore is milled down to 80-85% passing 135 pm. The finely ground ore goes
through a system of thickeners and acidulation tanks to destroy the contained carbonates.
The slurry is then preheated to a temperature of 165-175°C as it passes through splash
vessels. After that, the slurry is oxidized in the autoclaves for 40-60 minutes at a
temperature of between 215-220°C and a pressure of 2,900 kPa. The sulfide sulfur
oxidation rate is on the order of 90-92%. The following reactions take place in the
autoclaves:
2FeS2 + 02 + 4H+ = 2Fe2+ + 4S° + 2H20 (2.6)
11
|
Colorado School of Mines
|
4Fe2+ + 02 + 4H+ = 4Fe3+ + 2H20 (2.7)
2S° + 302 + 2H20 = 4H+ + 2S042" (2.8)
CaC03 + H2S04 = CaS04 + C02 + H20 (2.9)
SULFIDE PRE-TREATMENT AUTOCLAVING NEUTRALIZATION
Mill Cyclone Overflow CIRCUIT
Thickeners
Lime Lime
Splaeh
Heating Cooling
Towers Towers 2
Overflow to
Mill Reclaim
Water Tank
Oxygen
Steam
Autoclave #1
Splash
Heating
Towers
Oxygen,
Steam '
Autoclave #2 - #6
Figure 2.3. Autoclaving operation flowsheet (18)
This process demonstrates a few merits: a high efficiency of sulfide oxidation (90-
92%), a short retention time (40-60 minutes), and a practical absence of off-gases. The
primary disadvantages are high operating and capital costs.
2.3 Bio-oxidation Methods
The application of biotechnology to mineral processing has become feasible over
the past two decades (22). The bio-oxidation of metal ores can be defined as the leaching
of metal sulfides through the use of microorganisms. The role that microorganisms play
is very complex, but it can be said that microorganisms expedite chemical reactions.
Subchapter 2.3.1 addresses fundamentals and mechanisms of bio-oxidation of precious
metal ores. Subchapter 2.3.2 discusses applications of biotechnology and advances in this
field.
12
|
Colorado School of Mines
|
2.3.1 Theory of Bio-oxidative Pretreatment
Biological oxidation uses certain microorganisms to oxidize sulfide minerals and
can serve as either an ancillary or primary operation. For example, in the extraction of
copper, bio-oxidation can be utilized as a primary process to leach copper from both ores
and concentrates. Biological oxidation can also serve as a pretreatment process to liberate
precious metals occluded in sulfide minerals. This pretreatment process exposes precious
metals to chemical lixiviants.
Microorganisms can be classified into three groups, depending on the temperature
range at which they actively thrive, as presented in Table 2.1 (23-30).
Microorganisms catalyze two reactions to derive energy:
S0 (s) + 0 2 + 2 H20 4'H"' + s o / (2.10)
4-Fe2+ + 0 2 + 4 H+ 4-Fe3+ + 2-H20 (2.11)
Table 2.1. Groups of microorganisms used in bio-oxidation
Group Temperature range Major types
Mesophilic bacteria Acidithiobacillus ferrooxidans,
15-45°C Acidithiobacillus thiooxidans,
Leptospirillum ferrooxidans
Moderately Sulfobacillus thermosulfidooxidans,
thermophilic 40-65°C Sulfobacillus acidophilus, Acidophilus
bacteria ferrooxidans, Thiobacillus caldus
Extremely Sulfolobus acidocaldarius, Sulfolobus
thermophilic 60-95°C metallicus, Acidianus brierleyi
bacteria
Figure 2.4 represents two different pathways for leaching sulfides: the thiosulfate
and polysulfide pathways (31, 32). According to the thiosulfate pathway, sulfide minerals
are solely oxidized by ferric iron (Fe3+) as shown in Figure 2.4-A. This mechanism is
only relevant to acid-nonsoluble metal sulfides, such as pyrite (FeSi), molybdenite
(M0S2), and tungstenite (WS2). As a result of the first oxidation step, ferrous iron (Fe2+),
metal ion (M2+), and thiosulfate (S2O32 ) are generated. If sulfur-oxidizing bacteria are
absent, a significant amount of elemental sulfur might be produced. Microorganisms
regenerate ferric iron according to Reaction 2.11. The second mechanism is relevant to
13
|
Colorado School of Mines
|
Based on the literature review, Table 2.2 summarizes the primary factors affecting
bio-oxidation (16, 33-37).
Table 2.2. The key factors affecting bio-oxidation
FACTORS THRESHOLD DESCRIPTION
A consortium of Sulfur-oxidizing Some microorganisms oxidize iron, while some
microorganisms and iron-oxidizing bacteria oxidize only sulfur. It requires having both
(see Table 2.1) microorganisms types to maintain the optimal conditions of bio
are required for oxidation.
efficient bio
oxidation
Temperature (°C) Optimum Operating at an optimal temperature expedites the
temperature range kinetics of biooxidation.
depends on types
of microorganisms
(see Table 2.1).
PH The optimum Bacteria used in bio-oxidation are acidophilic. They
range is 1.4-2.0. can successfully thrive at the pH range of 1.2-2.5. If
pH exceeds 2.0, iron oxides start precipitating. These
oxides settle down on the pyrite’s (sulfides’) surface,
negatively influencing the bio-oxidation efficiency.
Oxygen (ppm) In the range of Bacteria need oxygen to proceed with bio-oxidation
2-5 ppm. reactions in the heaps and tanks. The optimum level
of dissolved oxygen is required to sustain the fast
kinetics.
Soluble cations and Individual. Many cations and anions have a toxic effect on
anions: As3+, Cu2+, Al3+, bacteria. Bacteria need to be acclimated to the ore
NO; and others before the process is developed.
Carbon dioxide (C02) More is better. Bacteria reduce carbon dioxide to build their cells so
it is a necessary component in the process. There are
two sources of carbon dioxide. The first source is air,
and the second is carbon dioxide evolving from the
acid neutralization reactions.
Redox potential Higher is better. One of the key factors in bio-oxidation is redox
potential, which is a driving force of pyrite leaching.
From a constant high redox potential, it can be
deduced that a rate of oxidation of ferrous iron into
ferric iron is enough to maintain the appropriate pyrite
leaching kinetics.
Nutrients: PO3 4, NH4+, Individual. There are some nutrients which are required for
Mg2+, K+ and others bacteria. They are quite often present in the ore. If
they are not, they need to be added.
Chloride ion (g/L) The maximum The chloride ion is toxic to bacteria. If saline water is
level is in the used, it slows down the growth of microorganisms.
range of Our own tests and available publications indicate that
15-20 g/L. bacteria can adapt to chloride ion concentration to
some extent.
15
|
Colorado School of Mines
|
2.3.2 Industrial Practice of Bio-oxidative Pretreatment
Several proprietary bioleaching technologies have been developed and
commercialized: the BIOX® process (38, 39), the BacTech process (40, 41), the BRGM
process (42), the GEOCOAT process (43-45), the BIOPRO process (4), and the HIOX®
process (46). Regardless of the technical approach used, from the perspective of
pretreatment, they can be divided into two groups for precious metal ores:
1. the stirred tank bio-oxidation of gold concentrates or ores;
2. the heap bio-oxidation of gold concentrates or ores.
The stirred tank bio-oxidation method is conducted in tanks on a finely ground
material. Figure 2.5 depicts a typical BIOX® process flowsheet, developed by GENCOR
S.A., Ltd. in South Africa. The typical size of feed concentrate would be approximately
P8o smaller than 75 pm with a solid content of 20% by weight. The pulp goes through a
series of tanks representing a two-stage oxidation circuit. The BIOX® bacterial culture
Thiobacillus ferrooxidans, Thiobacillus thiooxidans Leptospirillum
comprises and
ferrooxidans.
These microorganisms fall into the mesophilic group of bacteria that can
successfully thrive in the range of 30-45°C. Since bio-oxidative reactions are exothermic,
this process generates a significant amount of heat. Consequently, the bioreactors need to
be cooled in order to maintain an optimal temperature of 40-45°C (Figure 2.5).
Secondary Oxidation Tanks
y----
Primary Oxidation Tanks
Figure 2.5. Stirred tank bio-oxidation (38)
16
|
Colorado School of Mines
|
The company claims that the pulp residence time in the bio-oxidation plant is
approximately four days. After bio-oxidation is completed, the pulp goes through a three-
stage counter-current décantation circuit. The washed product goes to cyanidation, while
the final overflow liquor is neutralized by limestone and the precipitate is disposed of at a
tailings facility. There are six commercial operating BIOX® plants that have been built
since this technology was developed. Data from these plants indicate that gold recovery
might reach 90-95% after the BIOX® pretreatment. The company conducted a study to
estimate capital and operating costs, where the results showed considerable cost savings
over roasting and pressure oxidation (38).
Figures 2.6 and 2.7 illustrate the whole-ore heap bio-oxidation process
implemented in Nevada by the Newmont Mining Corporation (4). Newmont has
processed more than 8.8 million st of sulfidic ore and recovered more than 12.2 st of gold
using this pretreatment technology since its implementation in late 1999 (4). The
successful integration of the whole-ore heap bio-oxidation technology proved to be a less
capital-intensive alternative to other conventional processes and possessed lower
operating costs.
Figure 2.6 shows three bio-oxidation pads located on the mine site. Each heap is
designed to be 147 m wide and 305 m long. While one pad can accommodate
approximately 810,000 st of ore at a 10 m height, the heaps can reach a height as great as
16 m (4). All pads are aerated underneath by a piping network to provide oxygen and
carbon dioxide to the microorganisms. Figure 2.7 depicts a simplified diagram of the
process. The ore goes through a two-stage crushing circuit to liberate pyrite grains. After
crushing, sulfuric acid and inoculum are added to facilitate bio-oxidation. Trucks haul the
ore onto the pad and stack it in a heap. After that, the bio-oxidation cycle begins. Initially,
the bio-oxidized ore was designated to be followed by heap leaching for gold recovery.
However, due to excess capacity at the mill, once oxidation of the ore is completed, the
bio-oxidized ore is removed and hauled by trucks to the mill to be ground and processed
by the flotation and carbon-in-leach (CIL) circuits. The company reports 53.6% gold
recovery due to the bio-oxidation treatment (4).
While this technology has shown to possess a relatively low operating cost and is
less capital intensive, it has drawbacks associated with the process kinetics (one bio-
17
|
Colorado School of Mines
|
oxidation cycle takes 435 days) and gold recovery (53.6%). The length of the cycle and
the low metal recovery are the primary concerns associated with this process.
Ü g * g
Figure 2.6. Bio-oxidation pads in Nevada (4)
Two factors are believed to be responsible for these problems. The first factor
incorporates the conditions of habitat for bacterial growth. Some parameters, such as
oxygen flow through the heap and the surface of pyrite being exposed, are not at an
optimal level. The second factor is related to a crush size. The bio-oxidation efficiency is
a function of the size of ore particles (i.e., a smaller crush size tends to render a better
liberation of sulfides). As a result, a greater specific surface area of pyrite is exposed to
the biosolution. Dr. Brierley ran tests on this ore and showed that decreasing the crush
size from P8o=19 mm to Pgo=9 mm would improve gold recovery by about 10% (4).
However, it is economically impractical to crush the ore as fine as required in order to
achieve a maximum recovery. In addition, the ore stacked in the heap needs to be coarse
enough to allow for the biosolution and air to flow through the heap.
To overcome the drawbacks associated with the stirred tank bio-oxidation of gold
concentrates/ores as well as the whole-ore heap bio-oxidation, GeoBiotics, Inc.
developed the Geobiotics Process as shown in Figure 2.8 (43-45). Finely ground (212-38
pm) sulfidic gold-bearing concentrates are coated onto a support material, and these
18
|
Colorado School of Mines
|
agglomerates are then stacked into heaps as a part of the conventional whole-ore heap
bio-oxidation process. After bio-oxidation is completed, the agglomerates are removed
from the pad and screened to separate the support material from concentrate particles.
The undersize goes to a conventional CIL. According to the company, concentrates
containing as low as 15 g/t may be economically bio-oxidized (43-45). Experimental
results of column testing show that this technology requires a residence time of 50-100
days with an anticipated gold recovery of 80-95%.
BIOSOLUTION
TWO STAGE to
CRUSHING HEAP
ORE
from
STOCKPILE
SULFURIC
ACID
TRUCK HUAL BIO-HEAP TRUCK HUAL
to CELL to
BIO-HEAP CIL MILL
AERATION
FANS
INOCULUM
CRUSHING
BIO SOLUTION
POND
Figure 2.7. Whole-ore bio-oxidation technology in Nevada (4)
This pre-treatment process aims to incorporate advantages of the heap and stirred
tank bio-oxidation methods. Indeed, pretreatment is carried out on the pad that resembles
the whole-ore heap bio-oxidation. The fact that this technology uses sulfidic concentrates
presents an analogy between the Geobiotics Process and the stirred tank bio-oxidation of
gold concentrates. Needless to say, the gold recovery for the Geobiotics Process is higher
than the recovery for the whole-ore heap bio-oxidation. In addition, decreasing the
residence time translates to a shorter bio-oxidation cycle for the Geobiotics Process.
However, this process requires two ancillary operations, coating and screening
(Figure 2.8) which require additional capital and operating costs. Conversely, the
19
|
Colorado School of Mines
|
This table demonstrates seven important factors that need to be considered before
selecting a pretreatment method. As seen, each pretreatment method has advantages and
disadvantages. The whole-ore heap bio-oxidation demonstrates the slowest kinetics and
the poorest process efficiency. However, with systematic improvements in the process
efficiency and kinetics, the whole-ore heap bio-oxidation can potentially achieve a
competitive economic advantage over conventional roasting and oxidation.
The literature review of available bio-oxidation technologies has revealed that all
of the processes can be grouped into three categories, presented in Table 2.4. This table
demonstrates a seven-dimension qualitative comparison of these three categories. As
shown, the most efficient technologically is the stirred tank bio-oxidation. However, this
technology requires large capital and operating costs. The most economically desirable
method is the whole-ore heap bio-oxidation. This method, however, demonstrates the
poorest performance in terms of kinetics and recovery. This has caused some
investigators to express doubts about the commercial applicability of whole-ore heap bio
oxidation (1). The performances of GEOCOAT® stand somewhere between the previous
two technologies.
Table 2.4 A qualitative comparison of different bio-oxidation methods
Bio Economics Technological performances
oxidation Capital Operating Kinetics Gold Sensitivity Required Required
methods Cost Cost Recovery Head Reserves
Grade
Stirred tank Highest Highest Fastest High OK High Large
biooxidation
GEOCOAT® Moderate Moderate Moderate High OK High Moderate
Whole-ore Lowest Lowest Lowest Low Sensitive Low Large
heap
bio-oxidation
There is only one commercial whole-ore heap bio-oxidation operation known to
exist in the mining industry (4). Merits of this technology can be summarized as
possessing low capital and operating costs, and requiring a relatively short time to
commission the whole-ore heap bio-oxidation operation. Nevertheless, this method
currently has several significant disadvantages: a low gold recovery, a long retention
21
|
Colorado School of Mines
|
CHAPTER 3
ANALYTICAL PROCEDURES AND EXPERIMENT DESIGN
To achieve the research objectives, the following tests were conducted in the
Mineral Processing Laboratory at the CSM Mining Engineering Department:
1. Gold by Fire Assay and FAAS (flame atomic absorption spectrometry);
2. Iron by FAAS;
3. Free cyanide and protective alkalinity by titration;
4. Calcite equivalency by the Sobek procedure (47);
5. Total sulfur as sulfate by a gravimetric procedure.
Acidity (pH) and Reduction-Oxidation potential (pE) were measured in the Senior
Design Laboratory at the CSM Department of Metallurgical and Materials Engineering,
by using an electrochemistry meter (Model PH250) manufactured by Denver Instrument
Company.
Three commercial laboratories were engaged in this research for two reasons: to
validate obtained results and to conduct tests that CSM did not have equipment to
perform. These laboratories included:
• ACME Analytical Laboratories LTD;
• The Mineral Lab, Inc.;
• Jensen Technologies LLC.
All analytical procedures and experiments used in this research were scrutinized
to ensure the highest accuracy and to determine possible limitations. The research
program determined key statistics of analytical procedures and experiments as means,
standard deviations and relative errors.
3.1 Analytical Procedure for Gold Assay and Experimental Set-ups
To separate gold and silver from a geological material, a fire assay set-up was
assembled, shown in Figure 3.1. Buck Model 205 A AS, shown in Figure 3.2, was used to
determine the quantity of gold and silver extracted from the samples. Accurate fire
23
|
Colorado School of Mines
|
Table 3.3 contains the flux composition used to fuse the light fraction (<2,665 kg/m3).
The fraction possessing a density greater than 2,665 kg/m3 was fire assayed by using the
flux presented in Table 3.4.
Table 3.4. Flux for the fire assay of the heavy fraction
Substance Quantity [g]
-6.68+4 mm fraction -10+1 mm fraction
Sample 14.583 14.583
Soda 20 22.5
Borax 5 5.0
Litharge 62 62
Flour 0 0.6
Niter 1.51 0
Total flux 88.51 90.10
If there were any concerns about achieving a full extraction of gold from any
given sample that possessed a high pyrite content, then the sample was pretreated to
ensure reliable results: chemical decomposition and roasting were used as pretreatment
operations before fire assay.
Chemical decomposition is a wet pretreatment method. A representative sample
of 100% passing 74 pm weighing one assay ton (one AT =29.166 g) was placed into a
500-ml Erlenmeyer flask. De ionized water (20-25 ml) was added to prevent over-boiling
as a result of a quick oxidation. Next, 20 ml of concentrated nitric acid (69.6%) was
poured into the flask, and the flask was swirled. Then, 130 ml of nitric acid was slowly
added, and the flask was placed on a hot plate. The temperature was raised to 60-70°C,
and when red fumes ceased to be generated, an additional 150 ml of nitric acid was
added. After that, chemical decomposition was carried out at the elevated temperature of
65-75°C for 3 hours. After 3 hours, the pulp was brought to boiling for 10 to 15 minutes.
The solution was allowed to cool and then diluted with 100 ml of de-ionized water. The
flask was left overnight and then filtrated through the # 42 filter paper. The residue was
washed off thoroughly with de-ionized water and dried out. The residue was placed in the
26
|
Colorado School of Mines
|
3.2 Analytical Procedure for Analyzing Iron Leached by a Microbial Culture in
the Flasks
Erlenmeyer/bottom-baffled flasks of 500 ml volume containing samples were
placed into a water bath at 35°C, as shown in Figures 3.7. One flask labeled as a “blank”
consisted of pure quartz with the same weight as the other samples. This flask also had
the same volume and a composition of nutrient medium. The “blank” was used in order
to make adjustments for iron concentration by FAAS. Once a measurement was obtained,
the iron concentration [ppm] due to dissolution from the sample could be calculated from
the following equation:
(3.D
Xj X R is a Ro
where and are dilution factors; sample read-out [ppm]; is a blank read-out
2
[ppm].
Thus, the amount of iron that was dissolved from one mass unit of a sample could
be determined [%]:
_V ■Xr X2 (R-R0)
Fe M 10000
where M is sample mass [g], and V is volume of solution in the flask [ml].
Dilution factors were determined using a 10 ml pipette, and 250 ml and 500 ml
volumetric flasks. A 10 ml pipette was used to take a sample from each flask. The
quantity of sampled solution is very critical because it affects the accuracy of the test (i.e.
a larger sample yields better accuracy). However, it is not desirable to disturb the system
by taking too much sample. Another important factor is interference (i.e. a large sample
causes lower dilution, which might not be enough to eliminate or reduce the
interferences). After all these considerations had been taken into account, two possible
quantities of the sample were considered, 5 ml and 10 ml. For example, if a 10 ml sample
were taken from a flask and put into the flask containing 240 ml of de-ionized water.
28 ARTHUR LAKES LIBRARY
COLORADO SCHOOL OF MINES
GOLDEN, CO 80401
|
Colorado School of Mines
|
After that, a 5 ml sample were taken from this flask and diluted in the flask containing
495 ml of de-ionized water. The final dilution factor could be calculated:
x _
250-500 _ 2500.j
10 5
There is no description of this procedure in the literature addressing its accuracy
and reliability. A special research program was established to ensure that the quantity of
bio-oxidized iron determined using this method was valid and accurate.
Accuracy estimation requires identifying all operations involved in the
experiment, along with quantified estimates of inaccuracy that might be caused by each
operation. Accuracy is usually expressed by a distribution function, which is often a
normal distribution function and the standard deviation associated with it. Thus, the
V, X], X, R, Ro, M
standard deviations need to be defined for the variables in Equation
2
3.2. The standard deviations for each step are based on manufacturers’ data and this
study’s estimates are provided in Table 3.5. These data were analyzed using the @RISK
software package in order to execute the Monte Carlo simulation required to obtain
statistics for Equation 3.2 (49).
Table 3.5. Glassware and equipment characteristics
Item Standard Deviations for the Quantity Used
normal distribution
10 ml pipette 0.03 ml 5 ml and 10 ml
500 ml bottom-baffled flask 1 ml 400 ml
250 ml flask 0.17 250 ml
500 ml flask 0.33 500 ml
Mass balance 0.03 g 100 g
FAAS 0.1 ppm 0.3 ppm-5 ppm
Figures 3.4-3.5 represent examples of screenshots of simulated distributions for
iron concentration and percentage of bio-oxidized iron respectively. The important
statistics can be obtained from the figures such as mean, standard deviation, and the fifth
and ninety-fifth percentiles. The maximum error which might occur is calculated by
Equation 3.3:
29
|
Colorado School of Mines
|
E± =100 ^ ^ 5%/95% mean I (3.3)
mean
(Xs%)
Where the minus sign refers to the fifth percentile and the plus sign refers to the
(X95%).
ninety-fifth percentile
FAAS has a linear range read-out up to 5 ppm for iron. In fact, the level of
measurement can be regulated by varying a dilution factor in the lab. Consequently, the
relationship between read-outs and statistics characterizing the accuracy of iron
determination by the proposed experimental protocol needs to be determined. Figure 3.6
demonstrates this relationship. As seen, an increase in read-out, decreases the standard
deviation and maximum error. After the 1 ppm read-out, the maximum error becomes
less than 10%, which is acceptable for the most engineering projects.
The second issue is the sensitivity of the accuracy of the test to the quantity of
samples taken. There were two options being considered. Option A included taking a 5
ml sample and diluting it with 245 ml of de-ionized water. A 10 ml sample was then
taken from the later solution and diluted with 490 ml of de-ionized water. Option B
included taking a 10 ml sample and diluting it with 240 ml of de-ionized water. After
that, a 10 ml sample was taken from the later solution and diluted with 490 ml of de
ionized water. Table 3.6 presents the coefficient of variation at different levels of iron
bio-oxidation. As shown, carrying a bigger sample (10 ml in both steps) reduces the total
error of the experimental protocol by two times.
In conclusion, the accuracy valuation using the Monte Carlo simulation of the
experimental procedure for analyzing iron leached by a microbial culture in the flasks
revealed that this procedure could be used to determine iron concentration. After 10,000
iterations, the maximum error does not exceed 10%, which is statistically tolerable.
Table 3.6. Coefficient of variation at different iron levels
Iron grade simulated Coefficient of variation
% Option A (10-5) Option B (10-10)
0.50 9.44 4.72
1.49 • 3.20 1.65
2.97 1.73 . 0.91
30
|
Colorado School of Mines
|
experimental set-up for the “shaken flask” technique. A representative 10 kg sample was
crushed and pulverized down to 100% passing 74 pm. Then, representative 25 gram
samples were split and placed into 500 ml bottom-baffled flasks. The design of the
bottom-baffled flask increases the oxygen transfer coefficient, which is one of the crucial
factors in bio-oxidation. Each flask was filled with 250 ml of the nutrient medium and
1 ml inoculum was added into the flasks. The weights of the flasks were determined. One
flask, labeled as a “blank” and containing pure quartz with the same weight as the
samples, was included. This flask also contained the same volume and composition of
nutrient medium. The flasks were placed into the water bath, and the temperature was set
at 35°C. The hot shaker, shown in Figure 3.7, could accommodate up to 10 flasks. The
following parameters were monitored over time: pH, pE, and iron concentration in the
solution. After the bio-oxidation tests were completed, the iron content in the residues
was determined.
Figure 3.7. Hot shaker used for bio-oxidation tests
33
|
Colorado School of Mines
|
Figure 3.8 demonstrates the “bubbler bottle” technique. This technique was
utilized to grow bacteria on ferrous sulfate and to leach iron from coarse material of +1-
10 mm size. In order to bio-oxidize samples, the samples weighing 100 grams were
placed into 500-ml Erlenmeyer flasks, and the flasks were filled with 400 ml of the
nutrient medium, prepared as described previously. The “blank” flask was similarly
prepared. As before, all flasks were weighed and the weights were recorded. The flasks
were placed into a water bath, and the temperature was set at 35°C. Aquarium air pumps
(“Aqua Culture®”) delivered air into each flask at the rate of 1,200 cm3/min to maintain
the level of oxygen in the system. Before pH and pE were measured, the weights of the
flasks were checked by using a balance (“Sartorius”). If adjustments were required due to
evaporation losses, de-ionized water was added to bring the mass back to the original
weight. If some amount of the sample was taken from a flask to determine iron
concentration, the same amount of the nutrient medium was poured into the flasks to
account for iron loss and other losses from the system. To observe and count bacteria, a
phase contrast microscope (“SWIFT Series M950”) was utilized.
Acidity (pH) of the solution in which bacteria grow is one of the important factors
in bio-oxidation pre-treatment. Micro-organisms used in this research can thrive
successfully only in an acidic environment. This is why they are often referred to as
acidophilic bacteria. The literature review indicates that a range of pH from 1.2 to 3.5 is
considered suitable for these micro-organisms. Very low/high pH depresses microbial
activity, and in addition, high pH causes iron precipitation in the form of oxides (one of
them is jarosite) that lowers gold recovery due to the passivative effect on pyrite
oxidation (51, 52). The second negative effect from iron oxides is that they settle on the
gold’s surface and prevent gold from being exposed to the lixiviant, which results in low
gold recovery. For example, Figure 3.9 shows a residue that underwent bio-oxidative
treatment at pH of 2.2-2.5. The residue contains brown spots representing precipitated
iron oxides. Since the pH range given in the literature is extremely wide and the optimal
pH is dependent upon the material, the decision was made to run bio-oxidation tests by
using the “shaken flask” technique on the samples at different pH levels to determine the
optimum pH range for the studied system.
34
|
Colorado School of Mines
|
3.4 Experiments Design and Set-ups for Gravity Separation Study
The classic approach for testing gravity separation amenability, “float-sink”
analysis (FSA), uses “heavy liquids” (53, 54). The ore testing requires heavy liquids with
densities of more than 2,500 kg/m3. There are many types of liquids that can be applied to
ore separation - for example, methylene bromide (with a density of 2,490 kg/m3),
bromoform (with a density of 2,890 kg/m3), acetylene tetrabromide (with a density of
2,970 kg/m3), methylene iodide (with a density of 3,330 kg/m3) are common. Figure 3.11
depicts the procedure used for the heavy-liquid separation.
Depending on ore mineralogy, a few solutions are prepared using heavy liquid
and de-ionized or distilled water. The number of fractions that will be yielded exceeds the
number of made-up solutions by one. The required volume of the liquid can be calculated
from the following equation:
where Vo is required volume of the heavy liquid to be added [ml]; Vb is total volume of
the make-up solution in the beaker [ml] (specified); pi is required density [kg/m3]
(specified); po is density of the heavy liquid [kg/m3].
The solutions are prepared in opened vessels (usually beakers), as shown in
Figure 3.11. Densities are checked with a hydrometer before the testing begins. A sample
weighing a few kilograms is slowly fed into the beaker with the solution of the highest
density, and swirling by a rod is advised to ensure the perfect separation. Float fraction is
collected by a strainer and fed into the beaker with the solution having the next highest
density. The whole process continues until the last solution with the lowest density is
reached. All fractions need to be filtrated, rinsed, dried, weighed, and analyzed as
required.
Figure 3.12 shows a mineral jig that was used to carry out gravity separation tests.
The jig (“Sepor 9x16 Jig Concentrator”) is manufactured by Sepor, Inc.
37
|
Colorado School of Mines
|
Table 4.3. Results by ICP-MS
Me Mo Cu Pb Zn Ni As Cd Sb Bi Ag Au Hg TI Se
g/t 90.1 160.2 67.1 1574 320.8 2311.2 1.2 96.8 20.2 1.9 2.652 7.59 33.1 4.0
Table 4.4. Results by LiBO2/Li2B407 fusion + ICP-ES/MS
Item Si02 AI2O3 FezO] MgO CaO S C Ba
% 82.11 4.41 4.87 0.29 0.06 2.63 0.09 1.02
As observed, this analysis confirms the results obtained for quartz and pyrite
using the XRD and XRF methods. However, the amount of barium is higher (1.02%)
than the value determined by XRF. A possible explanation for this is that XRF is a semi-
quantitative method. If an assumption is made that the data from the ACME Analytical
Laboratories is more reliable, then the quantity of barite in the ore can be calculated as
1.73%. This figure complies with the result obtained using XRD. The small amount of
sulfur in the form of sulfate contained in the ore can then be calculated for barite and
equals to 0.24%.
An intensive test program was carried out by using FAAS and wet chemistry in
the CSM Mineral Processing Laboratory to determine an accurate measurement of gold
content. Table 4.5 represents results of the wet chemistry analysis.
Table 4.5. Wet chemistry analysis for gold
The level of confidence Confidence interval Average value Relative
for the Student’s (g/t) (g/t) error
distribution (%) (%)
95 0.28 2.73 10.16
Table 4.6 represents the results of the gold fire assay analysis.
Table 4.6. Fire assay analysis of gold
The level of confidence Confidence interval Average value Relative
for the Student’s (g/t) (g/t) error
distribution (%) (%)
95 0.06 2.61 2.39
42
|
Colorado School of Mines
|
Finally, this comprehensive investigation of the ore’s composition reveals the
following characteristics:
1. the ore contains a relatively low concentration of gold, where this material can be
accurately labeled as a low-grade ore;
2. the ore contains a sufficient quantity of trace elements (Mg and K) to maintain a
bacterial population;
3. the levels of metals such as Cu, As, Hg, Zn and A1 are not high enough to
suppress bacterial growth;
4. the ore has sufficient quantities of Fe and S to supply sources of energy for the
bacteria.
4.3 Direct Cyanide Leaching Study
A representative sample from the ore stock was pulverized down to (3-74=100%
and then prepared according to the process shown in Figure 4.3. The sample preparation
procedures aimed to obtain representative samples for direct cyanide leaching tests. The
portion of gold that can be leached without oxidative pre-treatment was calculated by
using Equation 4.1:
4 4
£ ~
100 ■ ,=l 4 , (4.1)
2> ,
1=1
where a; is gold content in the ore before the cyanide leaching test [g/t]; b, is gold content
in the residue after cyanide leaching tests [g/t].
Two samples of 70 g each, were placed into separate 500 ml PYREX beakers. A
sodium cyanide solution of 140 ml was used to leach each sample. The solution
composition was prepared as follows: sodium cyanide concentration (NaCN) = 1 g/L,
sodium hydroxide concentration (NaOH) = 1.1 g/L, pH of the solution = 11.5. The
samples were stirred by 2” stirring bars. The leaching tests were carried out over a 24
hour period.
43
|
Colorado School of Mines
|
The leaching tests revealed that the CLG after 24 hours was equal to 8.5%. As
such, this ore could be classified as a refractory ore that requires special pretreatment
before using direct cyanide leaching technology.
Titration tests were carried out to determine rates of sodium cyanide consumption
and protective alkalinity. The results of this work are presented in Table 4.7.
Table 4.7. Reagents consumption for cyanide leaching
Sodium cyanide used Sodium hydroxide used
kg/g of gold kg/t of ore g/L kg/g of gold kg/t of ore g/L
28.57 2.00 1.00 31.43 2.20 1.10
4.4 Bio-oxidation Study
Representative samples were crushed and pulverized down to 100% passing 74
pm. The 25-gram samples were then placed into 500-milliter bottom-baffled flasks filled
with a 250 ml culture medium prepared as described in Chapter 3. A bacterial culture,
previously grown on and acclimated to this ore, was inoculated. The flasks were
subsequently placed into the hot shaker’s water bath, and the temperature was set at 35°C.
As soon as bio-oxidation activity was noticed, one flask was taken out. The residue was
filtrated off, dried, and analyzed for iron. Then, the sample was leached by sodium
cyanide for 24 hours, and the residue was fire assayed to determine the gold content. As a
result of this process, two numbers were obtained: the percentage of iron remaining in the
residue after bio-oxidation and the gold content not recovered after bio-oxidation
followed by cyanide leaching. This procedure was repeated on the other flasks over a
period of seven weeks and the data were recorded. Figure 4.4 illustrates the plotted data
(i.e., the iron remaining in the residue [%] versus the gold remaining in the residue [g/t]).
As seen in Figure 4.4, there is a direct relationship between the iron removed from
the ore and the gold recovery by means of cyanide leaching. Therefore, this experiment
corroborates the premise of a tight association of gold with iron-bearing minerals. The
information contained in Figure 4.4 allows this relationship to be estimated. For example,
it can be calculated that 1% of iron in one tonne of ore locks up 0.71 g of gold. Based on
this ratio, a necessary level of bio-oxidation to obtain a desired technological
performance can be forecasted. Thus, experimental data strongly suggests that bio
45
|
Colorado School of Mines
|
CHAPTER 5
THEORETICAL CONSIDERATIONS
Solving any real-world problem requires the development of theoretical insight
into the phenomenon as a first step. Once all known factors and mechanisms driving the
phenomenon have been analyzed, it is then possible to formulate an approach which will
hopefully lead to a preliminary solution of the problem. This approach must then be
experimentally verified. Chapters 1 & 2 discuss the primary drawbacks of whole ore bio-
oxidative pretreatment. These chapters also focus on the objectives of this dissertation to
overcome these drawbacks as well as the theoretical and experimental tasks needed to
accomplish these objectives. This chapter aims to build an insightful observation of the
bio-oxidation phenomenon and to reveal a possible engineering solution to the challenges
addressed in the problem statement.
There are two closely related subsystems in bio-oxidation. The first subsystem
comprises “bio”, often defined as a microbial culture and constituents of the nutrient
medium that interact with ore particles. The second subsystem consists of ore particles
possessing a great variety of attributes. The second subsystem also responds to actions of
a microbial culture. As seen, both subsystems are mutually related. Therefore, a
comprehensive understanding of the phenomenon requires studying both subsystems
independently. First, a mathematical model of the pyrite dissolution mechanism promoted
by a microbial culture needs to be developed. Second, the characterization of individual
ore particles and their properties with respect to microbial activity needs to be
established. This approach results in a predictive theory that leads to a preliminary
solution for the stated challenges.
5.1 Mathematical Model of Pyrite Dissolution Mechanism Promoted by a Microbial
Culture
Any mathematical model incorporates a list of assumptions. The analysis
developed for this research contains the following assumptions:
1. The system contains pyrite, quartz, dolomite, and magnesite.
48
|
Colorado School of Mines
|
2. Microorganisms oxidize elemental sulfur and ferrous iron ions in order to gain
energy for metabolism.
3. The role that bacteria play is extremely complex, but specific generalizations can
be made. It can be assumed that bacteria increase rates of Reactions 5.1 and 5.2
by a catalytic mechanism. Reactions 5.3 and 5.4 are acid-consuming, and
Reaction 5.5 is electrochemical (31, 55).
2Fe2++ 0.5O2 + 2H+ -► 2Fe3+ + H20 (5.1)
S°+ 1.5-02 + H20 — 2 H+ + SOzf (5.2)
2 H+ + SO42" + CaC03 (s) -► CaS04 (s) + C02 (g) + H20 (5.3)
2 H+ + SO42 + MgC03 (s) -> MgS04 (s) + CO? (g) + H20 (5.4)
FeS2 (s) + 2-Fe3+ -> 3-Fe2+ + 2-S° (s) (5.5)
This model can be broken down into three parts, where each part can be
considered as an independent subsystem:
1) Acid consumption by alkaline minerals, such as calcium and magnesium
carbonates, is represented by Reactions 5.3 and 5.4. Available practical
experience and our own experimental data strongly indicate that this step takes
place relatively quickly and proceeds to the full destruction of all alkaline
minerals. If an ore is composed of a large percentage of carbonates, mining
companies usually acidify the ore with sulfuric acid before stacking the material
in heaps.
2) Species distribution in the solution is determined by equilibrium constants. The
system reaches the equilibrium relatively quickly because pyrite dissolution is a
rate-controlling reaction. Therefore, any additional amount of species introduced
into the solution immediately brings the system to a new equilibrium at any point
in time.
3) The kinetics of the process can be presented as a system of four equations
representing homogeneous and heterogeneous reactions: ferrous iron oxidation by
bacteria, pyrite dissolution, elemental sulfur production, and elemental sulfur
oxidation.
49
|
Colorado School of Mines
|
Ce).
the species distribution in the solution (i.e. Ci, C2, C3, C4, C5, This system was
solved by the method of substitution in MathCad.
Once a microbial culture is introduced in a nutrient medium, microorganisms start
oxidizing ferrous iron into ferric iron. This process leads to an increase in the C1/C2 ratio.
A system of equations (Equations 5.6-5.8 and 5.12-5.14) was used to determine how the
system responds to this bacterial activity. Simulations in MathCad revealed that the
ferrous iron oxidation increases the pH of the solution as shown in Figure 5.1.
Figure 5.1. The effect of the C1/C2 change due to bio-oxidation on pH
Figure 5.2 depicts the experimental results of changes in pH of the four different
samples in order to verify the theoretical conclusions obtained using the MathCad
simulation. As seen, the pH trend increases, decreases, and then stabilizes over time. This
behavior indicates that the first step, ferrous iron oxidation, consumes acid in the system
and consequently raises the pH (Reaction 5.1). The next step shows the generation of
sulfuric acid as a consequence of pyrite oxidation. This reaction results in a pH drop
(Reactions 5.2 and 5.5).
51
|
Colorado School of Mines
|
3.5
d, 2.5
1.5
0 2 4 6 8 10 12
Elapsed time [days]
Figure 5.2. Experimental results of pH change over time due to bio-oxidation of
four different materials: (1) the > 2,685 kg/m3 fraction; (2) ore; (3) the 2,600-2,685
kg/m3 fraction; and (4) the < 2,600 kg/m3
Available experimental evidence indicates that the first step in the bio-oxidation
process results from the attachment of bacteria on the surface of pyrite as illustrated in
Figure 5.3. Bacteria are scattered on the surface of pyrite, and they create a unique
environment in the vicinity of the pyrite surface with certain physical and chemical
properties (31, 32, 60-66). According to Reaction 5.1, bacteria convert two moles of
ferrous irons into two moles of ferric iron. One mole of oxygen accepts electrons from
ferrous iron and reacts with two moles of hydrogen to create one mole of water. Then, the
kinetics of ferrous iron oxidation into ferric iron promoted by bacteria (5.1) can be
written as:
^ = kr Sa-B-Cy Cf,
(5.15)
dt
52
|
Colorado School of Mines
|
where kg is rate constant [l/(s • m2 • cells/ml • M)]; Sois total surface area of pyrite [m2]; B
is the concentration of bacteria in the solution [cells/ml]; C9 is oxygen concentration (O2)
[M]; n2 is order of reaction.
The second process which occurs simultaneously with ferrous oxidation on the
surface is pyrite dissolution, represented by Reaction 5.1. This reaction is considered as
an electrochemical process. Any electrochemical reaction can happen only if anodic and
cathodic sites exist on the surface of a mineral (67). A visual representation of pyrite
dissolution is shown in Figure 5.4. The following reaction occurs on the anodic and
cathodic sites:
Anodic reaction: 2-Fe+++ + Fe++ + 2-e —> 3-Fe++ (5.16)
Cathodic reaction: 2-S~ - 2-e —> 2-S°(s). (5.17)
> \
Bacterium
0.50
•+
i+\
Pyrite (FeS2) 2H
.+
+/
0.50
Area influenced
2H
by bacteria
Figure 5.3. Attached bacteria on the surface of pyrite
Since pyrite is a semiconductor, electrons can flow from cathodic sites to anodic
sites. As such, ferric ions and elemental sulfur are generated on the surface of pyrite
minerals. The kinetics of the anodic reaction can be written as:
53
|
Colorado School of Mines
|
where A is a fraction of pyrite surface used as the cathodic site. This ratio indicates that a
constant rate of ferric iron production, bacterial concentration in the solution, and oxygen
concentration are important parameters that drive up the oxidation-reduction potential.
Indeed, in mining operations using the BIOX® process and the whole ore heap bio
oxidation, air is injected into the tank/heap to maintain the level of oxygen in the bio
solution. This process maintains oxygen-consuming reactions and the bacterial growth in
the system. Figure 5.7 shows experimental data of the red-ox potential versus time.
According to this data, after some period of time the red-ox potential levels off. This
observation complies with the theoretical conclusion presented in Figure 5.6. Thus, the
experimental data (Figure 5.7) corroborate conclusions drawn from the theoretical model.
This model can therefore serve as a predictive tool allowing for the identification of the
primary favorable factors affecting bio-oxidation.
3.5
2.5
u
u
0.5
0
2 4 6 8 10
elapsed time [days]
Figure 5.6. The C1/C2 ratio over time at different M/A ratios: (1) 0.2; (2) 0.7; (3) 2;
(4) 4; (5) 6; (6) 8; and (7) 10
From the obtained theoretical conclusions, it can be stated that a key factor for
efficient bio-oxidation is the high M/A ratio. To make it high, there are three factors
58
|
Colorado School of Mines
|
where |i is the specific growth rate [h1]; pmax is the maximum specific growth rate [h1];
Ks is the Monad constant; S is substrate [g/1].
As seen, there is a direct relationship between a substrate concentration and the
specific growth rate. Consequently, a system containing more substrate will demonstrate
the faster kinetics of bacterial growth and a higher final cell concentration in the solution.
If Monad’s equation is applied to the system studied in this research, a substrate (S)
represents iron and sulfur as sources of energy for microorganisms. Thus, an increase in
the surface area of pyrite will result in better conditions for bacterial growth that
expedites the pyrite oxidation.
5.2 Characterization of an Individual Ore particle’s Properties in Regard to a
Microbial Activity
For the purposes of this research, an ore is defined as a mixture of solid particles.
Under this premise, any pretreatment process is based on a specific property or a group of
properties of individual particles from this mixture. When ore particles are exposed to
any pretreatment process, a portion of ore particles acts differently from the other
particles according to their respective properties.. Thus, some particles respond to
different technological treatments and others do not. Obviously, to manage the process
efficiently, those favorable and unfavorable properties of an individual ore particle for
bio-oxidation need to be determined and thoroughly investigated.
In reality, there is always a combination of properties for any given particle that
determine liberation efficiency. Table 5.1 summarizes the primary characteristics of ore
particles with respect to a given sample.
60
|
Colorado School of Mines
|
Table 5.1. Characteristics of ore particle’s attributes
ATTRIBUTES THRESHOLD/REMARKS DESCRIPTION
(FACTORS)
Iron (Fe (II)) A rule of thumb is 1.5-2% Iron plays two important roles in
content [%]. pyrite (minimum). So the bio-oxidation. First, iron is a
minimum level of iron source of energy for bacteria (i.e.;
concentration in the ore is bacteria oxidize ferrous iron (Fe2+)
0.7-1%. Favorable factor. into ferric iron (Fe3+) to gain
energy). Second, ferric iron is a
strong oxidizer which oxidizes
pyrite and other sulfides.
Sulfur content The minimum level of It serves as a source of energy (i.e.,
(°S) [%]. concentration is 0.8-1%. bacteria oxidize elemental sulfur
Favorable factor. (°S) into sulfate ion (SO42 )).
Size of material. Top size depends on the The size reduction operation
liberation efficiency. intends to liberate sulfides so that
Generally, top size for bio their surfaces will be exposed to
oxidation in heaps may vary bacteria and the oxidizing solution.
from 15 to 60 mm. The size The second issue in heap bio
for bio-oxidation in tanks is oxidation is heap permeability to
around 65-90% passing 74 the solution and oxygen. Too many
pm. A bigger particle size is fine particles may prevent or
a favorable factor for the inhibit percolation and sufficient
whole ore bio-oxidation. oxygen access. In tank bio
oxidation, the apparent viscosity
increases as the size of the ore
decreases, which negatively affects
the oxygen transfer rate.
Specific surface Larger is better. This factor promotes the leaching
area of pyrite kinetics and provides bacteria a
(sulfides). surface for attachment.
Carbonates Less is better. Carbonates dissolve in the acidic
(CaCOs and solution consuming acid. As a
MgCOg). result, the pH might exceed the
optimum upper range.
Carbon (C). Less is better. Carbon-containing materials are
another source of “refractoriness”
called preg-robbing. They adsorb
gold on the surface preventing it
from dissolving in the cyanide
solution.
Gold content. More is better. Gold content affects the economic
efficiency.
61
|
Colorado School of Mines
|
It is seen that the first five attributes increase towards the black particle (heaviest)
comprising only pyrite; meanwhile, the sixth attribute increases towards the white
particle (lightest) comprising only quartz. Since, iron and sulfur are constituents of pyrite;
particles containing more pyrite will have more of those elements. The same logic can be
applied to the specific surface area of pyrite. The previous sections of this dissertation
showed that more than 90% of the gold is locked up in the pyrite. Therefore, it can be
concluded that particles with more pyrite contain more gold. However, carbon and
carbonates are relatively light substances, so their quantities rise towards the lightest
particle. Thus, particles with favorable attributes should have a higher density than
particles with unfavorable attributes. It is therefore possible to utilize the first attribute
(density) to separate ore particles by virtue of gold content.
Gravity separation (i.e., the method of separation based on differences in density
among the minerals) is probably the oldest technique used in Mineral Processing.
Nevertheless, it is still broadly used in industry as a primary or ancillary process for many
different ores and mineral commodities. From Figure 5.8, if a cut density is d^, it will
yield two different products: the light product (relatively high in carbonates and carbon
but low in sulfur, iron and gold); and the heavy product (relatively higher in iron, sulfur
and gold but low in carbon and carbonates). In addition, the specific surface area of pyrite
in the heavy product is expected to be larger, which enhances the kinetics of iron
oxidation. This being the case, only the heavy gravity concentrate that possesses the most
favorable attributes should be subjected to the bio-oxidation pretreatment. Meanwhile,
the light product possessing the least favorable attributes, should be directly processed in
the mill or stacked onto a heap for cyanide leaching without pretreatment.
63
|
Colorado School of Mines
|
CHAPTER 6
GRAVITY SEPARATION AMENABILITY STUDY
While Chapter 5 discussed the theoretical basis for using gravity separation
technology to improve the bio-oxidation pre-treatment, this chapter presents the
experimental verification and technical feasibility of using this technology with respect to
the research objectives. Subsection 6.1 addresses the sample preparation procedure,
Subsection 6.2 presents the results of “float-sink” analysis (ESA), and Subsection 6.3
discusses the results of mineral jigging.
6.1 The Crush Size Selection and Sample Preparation for Gravity Separation
The particle size distribution of samples to be processed is crucial in mineral
processing. It determines the minimum weight of a representative sample that needs to be
taken for testing, the efficiency of liberation and of bio-oxidation, and the type and setup
of equipment needed for gravity separation. Therefore, the size of ore that is suitable to
gravity separation and bio-oxidation needs to be determined prior to actual test work.
The company from which the samples were obtained currently reduces the size of
the ore to Pgo=19 mm. However, Dr. Brieley performed bio-oxidation tests on this ore
and showed that the ore should be brought down to Pgo=10 mm. This conclusion is based
upon empiric results that showed an additional 10% recovery of gold was possible (4). As
such, a gravity separation circuit will demonstrate better performance to some extent on
finer material due to a better liberation. In addition, available gravity separation
equipment at CSM could not process ore coarser than 10 mm. Thus, a decision was made
to crush the entire sample down to the top size of 10 mm.
After crushing the material, two size fractions (-6.68+4 mm and -10+1 mm) were
obtained for FSA. Reasons for choosing the size range of -6.68+4 mm were based on the
following:
1. Tests on this size range represent approximately a median value of the -10+1 mm
size range and was intended to verify the rationale presented in Chapter 5. In
Figure 5.8, the assumption is made that all particles have the same size.
64
|
Colorado School of Mines
|
2. It is widely held that results obtained on samples of narrow size ranges are more
representative than those obtained on the sample of wide range of the same mass
due to the segregation phenomenon;
3. It is convenient to run FSA on a narrow size distribution.
After preliminary tests on the size fractions of -6.68+4 mm were successfully
completed, a gravity separation study on the -10+1 mm size fraction was performed. This
series of tests was intended to verify conclusions drawn from the first series of tests on a
limited sample (the -6.68+4 mm fraction).
It is essential to take a representative ore sample for any test in the laboratory.
Figure 6.1 demonstrates a sample preparation process that the -10+1 mm sample
underwent prior to FSA (a similar procedure was conducted for the -6.68+4 mm sample).
A 28.9 kg sample was taken from the ore stock by coning and quartering. The sample
went through stage crushing, using a jaw crusher set at a gap size of 7-8 mm and a sieve
of 10 mm, to obtain a crushed material of 100% passing the 10 mm sieve. The sample
was then split by a Jones splitter and each sub-sample went through a series of four
splitting operations. Each half, except ones from the fourth splitting, were placed back in
the initial bucket. As a result of this procedure, four equal samples were generated. The
samples were screened on a 1 mm size sieve to obtain two fractions: -10+1 mm and -1+0
mm. This procedure rendered four equal samples of each fraction. All samples of the
-1+0 mm fraction were mixed and sent for further tests: fire assay, cyanide leaching, ICP-
MS, XRF, and XRD. Two samples of the -10+1 mm fraction were sent to FSA, one
sample was kept as a referee sample, and the last one was chemically analyzed.
65
|
Colorado School of Mines
|
6.2 FSA on the -6.68+4 mm Size Sample
A representative sample was obtained from the ore stock to conduct preliminary
test work on gravity concentration. The sample was screened to obtain the -6.68+4 mm
size fraction. Two media, with densities of 2,600 kg/m3 and 2,685 kg/m3, were prepared
from sodium polytungstate of 2,890 kg/m3. Table 6.1 presents the results of FSA.
Table 6.1. FSA on the -6.68+4 mm size fraction
Density Yield Gold Gold distribution Iron distribution
[kg/m3] [%] [g/t] [%] [%]
<2,600 9.52 1.04 4.51 8.58
2,600-2,685 70.30 0.83 26.55 60.09
>2,685 20.18 7.51 68.95 31.34
Total 100.00 2.20* 100.00 100.00
* - as weighted average. Low value of head grade due to a partial gold recovery from the heaviest fraction.
Based on the data presented in Table 6.1, Recovery-Yield curves for iron and gold
were plotted. Figure 6.2 depicts three curves. Curve #1 represents no gravity
processability and serves as a benchmark. The dash line (Curve #1) demonstrates that
recovery is directly proportional to yield, which means that employing gravity separation
is not feasible. The size of the area between the dash line and the other curves can serve
as a benchmark for a relative estimate of gravity separation amenability (i.e., a curve that
is close to the dash line shows less favorable conditions for processing by gravity
separation than a curve that is located far from the dash line). Curves #2 and #3 represent
gravity processability cases for iron and gold, respectively. The positions of these curves
for gold and iron indicate that this ore responds to gravity separation.
Table 6.1 shows that the <2,600 kg/m3 and 2,600-2,685 kg/m3 fractions contain almost
the same amount of iron, meaning that iron-bearing minerals are distributed evenly
among those fractions. In addition, a decrease in the iron content in those fractions is not
significant in comparison with the head grade of iron. In other words, gravity separation
does not result in the high enrichment ratio for these fractions. In contrast, the >2,685
kg/m3 fraction comprises a relatively high amount of iron.
67
|
Colorado School of Mines
|
Curve #2 is located closer to Curve #1 than to Curve #3; indicating that iron-
bearing minerals have less amenability to gravity separation in comparison with gold.
However, it has been established previously that gold is primarily occluded in iron
bearing minerals. There is also a small amount of free gold in this ore. To understand the
difference in the curves’ behavior, the ratios of the quantities of gold to iron in each
fraction needed to be analyzed. Under the premise that all gold is occluded in iron-
bearing minerals, it can be shown from Table 6.1 that the ratios for different fractions are
the following: 4.51/8.58 equals 0.52 for <2,600 kg/m3; 26.55/60.09 equals 0.44 for
2,600-2,685 kg/m3 and 68.95/31.34 equals 2.20 for >2,685 kg/m3. Therefore, it can be
concluded that iron-bearing minerals of the heaviest fraction are enriched with gold, in
comparison with those in the other fractions. In other words, gold might not be
distributed evenly among iron-bearing minerals. This suggests that there might be two
types of iron sulfides in the ore. The first type does not contain a significant amount of
gold, but the second one does. Therefore, two different responses to any technological
treatment should be considered for this ore. This series of tests lead to the following
conclusions:
1. The >2,685 kg/m3 fraction contains 4-5 times as much gold as the other fractions
in concentration of one percent of iron per tonne of material. This means that the
oxidation of one percent of iron in this fraction releases more gold than does one
percent of iron in the other fractions.
2. The >2,685 kg/m3 fraction contains iron and sulfur, which are favorable
substances to bacterial growth (almost two times as much as the lighter fractions
and initial sample).
3. All particles have the same size of -6.68+4 mm. This implies that every particle
from the heaviest fraction has more pyrite than particles from the other fractions
and the raw ore itself. This results in a bigger specific surface area of pyrite;
consequently, that leads to better kinetics of bio-oxidation.
4. Finally, the >2,685 kg/m3 fraction contains approximately 70% of total gold while
contributing only to 20.18% of the ore. This means that a bio-oxidation heap will
have more gold per tonne of ore stacked in it, and the capacity of the bio
oxidation operation can be increased significantly.
70
|
Colorado School of Mines
|
6.3 FSA on the -10+1 mm Size Sample
A representative sample of approximately 40 kg was obtained from the ore stock.
The sample was stage crushed to obtain a size of 100% passing 10 mm. The sample was
then homogenized and processed as shown in Figure 6.1. Figure 6.4 presents the result of
a particle size analysis performed on this material. As seen, the crushed material contains
9.35% of the 0-1 mm size fraction. This indicates that comminution produces the
oversize accounting for approximately 90.65% of the feed. The sample was screened on a
1 mm sieve to obtain the -10+1 mm size and the -1+0 size fractions. The -10+1 mm size
fraction was subjected to FSA in order to test gravity separation amenability for the wide
size range. This experimental work is described on the flowsheet in Figure 6.5.
Table 6.2 shows the results of FSA on the -10+1 mm fraction. As seen, the float
fraction (<2,665 kg/m3) represents 80.96% of the feed and contains 41.91% of total gold.
In contrast, the sink fraction (>2,665 kg/m3) accounts for 19.04% of the feed and collects
58.09% of total gold. Based on this analysis, the ore can be processed by means of
gravity separation.
Table 6.2. Results of FSA on the -10 +1 mm sample
Cumulative float Cumulative Cumulative
Density yield Gold gold iron
[kg/m3] [%] [g/t] recovery recovery
[%] [%]
2,500 0.00 0.00 0.00 0.00
2,610 25.08 1.62 15.93 18.17
2,665 80.96 1.32 41.91 69.84
2,762.5 100.00 2.55 100.00 100.00
71
|
Colorado School of Mines
|
The samples representing the <2665 kg/m3 and <2665 kg/m3 fractions were sent
to ACME Analytical Laboratories LTD and The Mineral Lab, Inc. to conduct analytical
work. Tables 6.3 and 6.4 present some of the results of this work (Appendices D and E).
Table 6.3. Results of analytical work on the -10 +1 mm sample from
ACME Analytical Laboratories LTD
Density Gold* Iron** Total sulfur
[kg/m3] [g/t] m [%]
4A/4.6
<2665 1.32 0.77/0.86***
>2665 7.75 10.05/9.76 6.18/7.05***
Total 2.55 5.53 2.04
* values obtained at CSM using the fire-assay method;
** the first value by leaching and analyzed by ICP-MS, the second value by LIB02/LI2B407
fusion and analyzed by ICP-ES;
*** total sulfur using LECO.
Table 6.4. Results of XRD for the -10 +1 mm sample from The Mineral Lab, Inc.
>2665 kg/m <2665 kg/m3
Minerals [%] [%]
Quartz 73 83
Pyrite 8 <2
As Table 6.3 indicates, the >2,665 kg/m3 fraction is significantly enriched by iron
and sulfur, favorable substances for microorganisms. As a result, better conditions for
bacterial growth should be expected with the >2,665 kg/m3 fraction. In addition, this
fraction contains 5.9 times more gold than the <2,665 kg/m3 fraction does.
Figures 6.6 and 6.7 depict the Recovery-Yield and the Yield/Distribution curves
of the float fractions. All curves are similar to those obtained on the narrow size range.
The shape of curves indicates that this ore can be processed by using gravity separation
technology.
74
|
Colorado School of Mines
|
6.4 Ore Jigging Study
There are several gravity separation technologies available to separate coarse
particles. Generally, the following rules are applicable to gravity separation technology
(70):
1. the size range of -100+10 mm is suitable for heavy media separators;
2. the size range of -40+1 mm is suitable for mineral jigs and heavy media cyclones;
3. the size range of -1+0.1 mm is suitable for spirals, shaking tables, and
hydrocyclones.
Given the physical limitations of the equipment available, a decision was made to
process the size of -10+1 mm using a jig. The entire sample (442 kg) was stage crushed
down to 100% passing 10 mm by using a jaw crusher. The entire sample was then
screened on 1 mm and 6.3 mm sieves using a “Gilson.” Figure 6.8 represents the
cumulative undersize distribution of the crushed sample. It can be seen that
the +1-10 mm fraction, which is subject to jigging, represents approximately 90.01% of
the feed. However, the design of the jig located at the Mining Engineering Department is
not suitable to process crushed ores finer than 6.3 mm. In addition, it is a common
practice to pre-classify an ore before jigging. Therefore, only the +6.3-10 mm fraction
representing approximately 37.06% of the crushed material underwent jigging. Results of
jigging on the +6.3-10 mm fraction can be extrapolated to the +1-6.3 fraction. Indeed,
separation tests on the +1-6.3 fraction would yield better results due to the better
liberation of sulfides.
Figures 6.9 and 6.10 represent the simulated flowsheets. The first flowsheet
(Figure 6.9) depicts the one stage jigging. Gravity separation produces a concentrate and
tailings. The concentrate has a yield of 39.19% with a gold recovery of 52.55%. The
tailings have a yield of 50.82% with a gold recovery of 35.63%. The second flowsheet
represents two stage jigging. This flowsheet depicts a concentrate with a yield of 24.36%
and a gold recovery of 43.49%, the middlings with a yield of 12.06% and a gold recovery
of 9.70%, and the tailings with a yield of 53.59% and a gold recovery of 34.90%.
77
|
Colorado School of Mines
|
pattern indicates that pyrite is being oxidized by ferric ions as described in Reaction 5.5
and that sulfuric acid is being generated according to Reaction 5.2.
Figure 7.2 demonstrates the pE change over a period of the bio-oxidation time. As
seen in the figure, pE increases rapidly and stabilizes at approximately 700 mV for the
sample representing the feed. In contrast, pE increases slightly and then stabilizes at
approximately 400 mV over a period of 40 days for the 2,600-2,685 kg/m3 fraction. After
40 days, pE increases sharply and stabilizes near 560 mV. At the end of the experiment,
cell concentrations were counted in each solution using a phase microscope. The solution
750
700
650 27x1 3 cell Si'mt
600
£7 550
ü
a 500
s/ml
450
400
350
300
0 10 20 30 40 50 60 70 80 90
Elapsed time |[days]
Figure 7.2. The pE change over a period of time: (1) feed; (2) the 2,600-2,685
kg/m3 fraction
accommodating the feed contained 27xl06 cells/ml, while the solution accommodating
the 2,600-2,685 kg/m3 fraction contained 12xl06 cells/ml. As shown, microorganisms
grown on the sample representing the feed have increased in the cell concentration two
times as much as microorganisms grown on the 2,600-2,685 kg/m3 fraction. This increase
indicates that bacteria prefer the feed rather than the 2,600-2,685 kg/m3 fraction.
83
|
Colorado School of Mines
|
700
.X"
600 7 0
4^ / / / /0
\
A
#X
i500 0
// /
t0 *
-
/
400 ----- y /
a
y''N
300
------^
200
+-4T
100
10 15 20 25
Elapsed time [days]
Figure 7.5. The pE change over a period of time: (1) the 2,600-2,685 kg/m3 fraction;
(2) feed; (3) the >2,685 kg/m3 fraction; (4) blank
approximately 660 mV after 13 days. Conversely, the 2,600-2,685 kg/m3 fraction
demonstrates relatively slow kinetics: pE reaches approximately 650 mV after 23 days.
These data indirectly indicate that microorganisms grow slowly in the solution hosting
the 2,600-2,685 kg/m3 fraction.
To conduct the bio-oxidation study on the coarse size range, three representative
samples of 100 grams were obtained and the “bubbler bottle” technique was conducted
on the samples. The samples of the -10+1 mm size were not pulverized for this test. They
represent the <2,665 kg/m3 fraction, the >2,665 kg/m3 fraction, and the -10+1 mm size
ore (feed) that did not undergo gravity separation. All samples were placed in 500-ml
flasks. The flasks were then filled with 400 ml of the nutrient medium prepared as
previously described in Table 3.7. The flasks were partially immersed into a water bath
and the temperature was set at 35°C. To maintain an appropriate level of oxygen in the
system, an aquarium pump delivered air into each solution at the rate of 1,200 cm3/min.
The intent of this series of tests was to make microorganisms acclimated to the samples.
86
|
Colorado School of Mines
|
As seen in Figure 7.7, pH increases to a maximum value, declines in value and
then stabilizes. This pattern is typical for bio-oxidation, where microorganisms first
promote acid-consuming reactions and then pyrite oxidation takes place, which translates
into a pH drop. As shown, the solution containing the >2,665 kg/m3 fraction has the
2
1.95
1.9
1.85
1.8
S 1.75
1.7
1.65
1.6
1.55
1.5
0 10 20 30 40 50 60 70 80 90
Elapsed time [days]
Figure 7.7. The pH change over a period of time: (1) feed; (2) the <2,665 kg/m3
fraction; (3) the >2,665 kg/m3 fraction
minimum value of pH, while the solution containing the <2,665 kg/m3 fraction has the
maximum value of pH after 85 days of the experiment. The solution containing the feed
has the intermediate pH. From these experimental data, and considering Reaction 5.2, a
conclusion can be drawn that the sample containing the >2,665 kg/m3 fraction generates
the greatest amount of sulfuric acid among the samples. To generate sulfuric acid,
elemental sulfur needs to be supplied to the system. The only source of elemental sulfur
in the flask is the product of Reaction 5.5, which is the pyrite oxidation reaction. This
implies that the greatest quantity of oxidized pyrite is found in the flask containing the
>2,665 kg/m3 fraction.
88
|
Colorado School of Mines
|
Red-ox potential behaves consistently with bacterial activity as well. It rises very
quickly for the >2,665 kg/m3 fraction and feed, and then it stabilizes at approximately 720
mV. Despite the rise of pE for the <2,665 kg/m3 fraction, the rate is not as high as the rate
for the >2,665 kg/m3 fraction and feed. In addition, the final pE is in the vicinity of 600
mV.
750
700 * /
650
X y
/ X
3
o
600
<
| 550 h
2
V s '’
% 500 r>!à
X<^_-
450
I z
400
350
300
10 20 30 40 50 60 70 80 90
Elapsed time [days]
Figure 7.8. The pE change over a period of time: (1) feed; (2) the <2,665
kg/m3 fraction; and (3) the >2,665 kg/m3 fraction
Figure 7.9 represents the pyrite leaching kinetics for all three samples. The graphs
show that one mass unit of the >2,665 kg/m3 fraction contains more than 3.58% leachable
iron by bacteria, the feed contains less than 1 % leachable iron by bacteria, and the <2,665
kg/m3 fraction contains 0.24% leachable iron by bacteria.
To conclude, the experimental data indicated that the bacteria became acclimated
to the >2,665 kg/m3 fraction and the feed sample, and successfully started leaching pyrite
in these samples. However, microorganisms in the flask containing the <2,665 kg/m3
fraction did not leach iron appreciably.
89
|
Colorado School of Mines
|
be estimated: 100 • (8.18+7.66) / (2 • 8) = 99%. Thus, the >2,665 kg/m3 fraction can be
efficiently oxidized by microorganisms.
Table 7.1. Summary of bio-oxidation on various samples of the -10+1 mm size
Sample Leachable iron Leachable pyrite
[%] [%]
value average value average
>2,665 kg/m3 #1 3.82 3.70 8.18 7.92
>2,665 kg/m3 #2 3.58 7.67
<2,665 kg/m3 #1 0.09 0.16 0.19 0.35
<2,665 kg/m3 #2 0.24 0.51
Feed #1 1.31 1.14 2.80 2.43
Feed #2 0.96 2.06
Based on the obtained data presented in Tables 7.1 and 6.2, the efficiency of bio
oxidation can be defined. Considering the fact that the >2,665 kg/m3 fraction represents
19.04% of the feed and contains 7.92% of the leachable pyrite, then the following amount
of pyrite would be leached from one mass unit of the feed if only the >2,665 kg/m3
fraction were stacked onto the bio-oxidation pads:
1 t • 0.1904 • 0.0792 = 0.01508 t.
To compare this efficiency with the efficiency of bio-oxidation of the feed, the equivalent
amount of the feed releasing the same amount of pyrite needs to be found as
X -0.0243 = 0.01508 t,
where X represents the quantity of the feed releasing 0.01508 t of pyrite. From this
equation, X equals 0.62 t. Therefore, 0.19 t of the >2,665 kg/m3 fraction is equivalent to
0.62 t of the feed in terms of the leached amount of pyrite. As such, if the bio-oxidation
operation aims to oxidize the same amount of pyrite over a period of time, an amount of
91
|
Colorado School of Mines
|
CHAPTER 8
CONCLUSIONS AND RECOMMENDATIONS FOR FUTURE STUDIES
8.1 Conclusions
This dissertation aims to improve the process efficiency associated with the heap
bio-oxidation of refractory gold ores through the development of a new design
methodology. The key concept of this methodology is based upon the premise that for
any crushed ore there exist particles possessing favorable and unfavorable characteristics
for bio-oxidative pretreatment. If only those particles conducive to bio-oxidation are
extracted from the crushed ore and then subsequently subjected to bio-oxidation, this
“selective” pretreatment will result in a better performance and process efficiency. To
verify this idea, a comprehensive research program has been completed including the
following steps:
• analysis of existing commercial operations;
• analysis of the state of the application of bio-technology to Mineral Processing;
• ore identification study;
• theoretical study;
• mathematical modeling;
• gravity separation amenability study;
• bio-oxidation amenability on various gravity fractions;
• mineral jigging amenability study.
This research led to the following major conclusions:
1. A microbial culture requires certain conditions for successful growth. The
presence of sufficient quantities of pyrite is one of the main conditions because
the oxidization of ferrous iron and sulfur is the sole source of energy for
microorganisms in the system studied. In addition, to initiate pyrite oxidation the
microorganisms must adhere onto the surface of a pyrite particle. This enables
them to form a special environment on this surface by creating areas serving as
centers of ferrous iron oxidation and pyrite oxidation.
94 ARTHUR LAKES LIBRARY
^ COLORADO SCHOOL OF MINES
GOLDEN, CO 80401
|
Colorado School of Mines
|
2. Mathematical modeling led to a conclusion that there are three factors affecting
the kinetics of pyrite leaching. They include the surface area of pyrite, microbial
concentration, and oxygen concentration. Simulation revealed that increasing
these factors expedites the kinetics of pyrite leaching.
3. Detailed investigation of an individual particle’s properties showed that there are
six main properties affecting bio-oxidation. Those properties can be classified into
two groups. The first group comprises favorable properties: iron content, sulfur
content, gold content, and pyrite specific surface area. In contrast, the second
group comprises unfavorable properties: carbon and carbonates contents.
Moreover, particles possessing favorable parameters have a higher density than
particles with unfavorable parameters. This creates an opportunity for using
gravity separation to separate particles conducive to bio-oxidation from particles
that are not.
4. The heavy fraction (>2,665 kg/m3) contains 5.9 times more gold than the light
fraction (<2,665 kg/m3) and 3 times more than the feed. In addition, the heavy
fraction contains iron and sulfur, favorable substances for microorganisms, at a
concentration of almost two times greater than the light fraction and the feed.
Another positive outcome of gravity separation is that the heavy fraction contains
less carbonates and carbon, unfavorable substances for microorganisms, than the
light fraction and the feed. As demonstrated, gravity separation reduces quantities
of unfavorable substances for bio-oxidation in the material. This results in faster
kinetics of bio-oxidation. Therefore, better conditions for bacterial growth are
created within the heavy fraction than either the feed or the light fraction.
5. A series of mineral jigging tests showed that a gravity concentrate with a weight
of 24.36% and a gold recovery of 43.49% could be obtained. The results obtained
through experimentation indicated that this ore could be processed by means of
gravity separation. However, the jig’s design limitations and the limited sample
size did not allow for further optimization of the process. The research supports
the conclusion that the results of gravity separation could be improved
significantly by using a larger sample and a jig designated for sulfides. Another
95
|
Colorado School of Mines
|
process alternative for preseparation of this ore is the use of heavy-media
centrifugal separation.
6. Table 8.1 contains the experimental results of bio-oxidation of various materials,
followed by gold leaching. As shown in the table, the bio-oxidized heavy fraction
(>2,665 kg/m3) experienced extremely improved rates of gold extraction.
Table 8.1. Summary of bio-oxidation and leaching studies
Item Gold Pyrite Leached Gold recovery
[g/t] [%] pyrite [%]
[%]
ROM ore 2.61 4-5 0 <9
Ore (-1+0 mm) 3.11 4-5 0 <9
Feed (-10+1 mm) 2.55 4-5 0 <9
Feed (-10+1 mm) 2.55 4-5 2.43 50-60
Jig tailings 1.7 <1 0 23-25
The <2,665 kg/m3 fraction 1.32 <1 0 <30
The >2,665 kg/m3 fraction 7.75 8-10 7.92 60-70
7. Since the weight of the >2,665 kg/m3 fraction is 19.04% and this fraction contains
7.92% of the leachable pyrite, 0.1904 t of the >2,665 kg/m3 fraction would render
15.08 kg of pyrite into the solution after bio-oxidation. The same amount of pyrite
could be produced from 0.62 t of the ROM ore. As such, if the bio-oxidation
operation oxidizes the same amount of pyrite over a period of time, the amount of
material stacked in heaps can be decreased by 3.26 times if the gravity separation
technology is employed. However, if the tonnage of bio-oxidation heaps is
maintained at the same level, the amount of pyrite released by the >2,665 kg/m3
fraction is as much as 3.26 times the amount of pyrite released by the ROM ore.
As a result, bio-oxidation becomes a much more efficient operation in terms of
tonnage of ore processed and the quantity of gold produced by a given heap pad.
96
|
Colorado School of Mines
|
8.2 Recommendations for Future Studies
Full scale gravity separation tests on the crushed ore of -10+1 mm size weighing
15-20 t need to be conducted. From past experience, two options might be considered for
this ore: a mineral jig and a two-stage Tri-Flo dynamic dense-medium separator. Upon
completion of the tests, a final decision can be made whether jigging or dense-medium
separation is the best option.
The developed methodology reveals many other alternatives for the processing of
products created by gravity separation. Figure 8.2 demonstrates the possible routes that
can be explored. A full line in Figure 8.2 depicts mandatory streams of material, while a
dash line depicts possible paths of material. The product of the heap bio-oxidation after
neutralization by lime can be leached in heaps for gold extraction. The fine concentrate
created by spiral concentrators can be further processed by employing the following
options: the GEOCOAT® process, the flotation process, and the oxidative roasting. The
coarse tailings from the gravity separation circuit may be either stacked in heaps and then
leached or sent to a flotation mill. Needless to say, some portion of these tailings can be
rejected as a coarse waste (used as a gravel material). Finally, the tailings created by
spiral concentrators might be processed by either a gold leaching circuit or a flotation
mill.
Since gold is encapsulated in the pyrite matrix, the material after bio-oxidation
needs to be finely ground prior to gold extraction. The effect of grinding on gold recovery
needs to be studied.
99
|
Colorado School of Mines
|
APPENDIX A
XRD REPORT
October 25, 2006
Lab no. 206715
Mr. Michael P. Thomas
M.H.S. Research
5770 West 4th Avenue
Lakewood, Colorado 80226
Dear Mr. Thomas:
Enclosed are the x-ray diffraction (XRD) results for your sample, “Aissautov" received last week.
A representative portion of the sample was ground to approximately -400 mesh in a steel swing
mill, packed into a well-type plastic holder and then scanned with the diffractometer over the
range, 3-61° 28 using Cu-Ka radiation. The results of the scan are summarized as approximate
mineral weight percents on the enclosed table. Estimates of mineral concentrations were made
using our XRF-determined elemental composition and the relative peak heights/areas on the XRD
scan. The detection limit for an average mineral in this sample is -1-3% and the analytical
reproducibility is approximately equal to the square root of the amount. "Unidentified" accounts
for that portion of the XRD scan which could not be resolved and a indicates doubt in both
mineral identification and amount.
Thank you for the opportunity to be of service to M.H.S. Research.
Peggy Dalheim
107
|
Colorado School of Mines
|
APPENDIX E
XRD REPORT
August 2, 2007
Lab no. 207560
Mr. Adil M. Aissautov
Mining Engineering Department
Colorado School of Mines
1600 Illinois Street
Golden, Colorado 80401
Dear Mr. Aissautov:
Enclosed are the x-ray diffraction (XRD) results for the four samples submitted last week. Also enclosed
are copies of the raw XRD data, as requested.
A representative portion of each sample was ground to approximately -400 mesh in a steel swing mill,
packed into a welMype plastic holder and then scanned with the diffractometer over the range, 3-61 "28
using Ct>Ka radiation. The results of the scans are summarized as approximate mineral weight percents
on the endosed table Estimates of mineral concentrations were made using our XRF-determined
elemental compositions and the relative peak heights/areas on the XRD scans. The detection limit for an
average mineral in these samples is ~ 1 -3% and the analytical reprodutibiiity is approximately equal to the
square root of the amount. "Unidentified" accounts for that portion of the XRD scan which could not be
resolved and a ”?” indicates doubt in both mineral identification and amount.
Thank you for the opportunity to be of service to CSM.
Peggy Dalheim
117
|
Colorado School of Mines
|
ABSTRACT
Thermal coal is a prominent resource from which electricity is produced. In recent
years, the price of this widely used commodity has declined, largely due to increases in envi-
ronmental regulations, incentives for renewable resources, and technological advances in the
production of natural gas, which is a cleaner-burning alternative fuel. As such, coal markets
have drastically changed. We develop optimization models to help understand the current
climate on a global scale and to help a domestic utility reevaluate its forward purchase strat-
egy given depressed coal prices. We first develop a global thermal coal optimization model
that minimize the cost to ship coal from countries that are net exporters to those that are net
importers, taking into account coal quality specifications and shipping constraints, such as
port and vessel capacity. Using this model, we explore the global effects of price variability,
changes in operation from large market players such as China and India, and the impact of
the Panama Canal expansion. We next develop a methodology for determining an optimal
purchase strategy for a U.S. utility; using historical observations, we build a regression model
to forecast prices, then select representative scenarios to include in a multi-stage stochastic
program that minimizes the expected value and conditional value-at-risk (CVaR) of coal
procurement. We formulate a time-consistent nested CVaR minimization model, compare
its performance to an expected CVaR model, and show that the expected CVaR model
may be better suited to minimizing risk in a multi-stage setting. We conduct out-of-sample
testing to assess our solution performance under new price realizations. Finally, we apply
an expected CVaR model to determine a utility’s procurement strategy and recommend a
purchase plan for implementation that is expected to save the company $151 million over a
five-year horizon.
iii
|
Colorado School of Mines
|
CHAPTER 1
INTRODUCTION
Thermal coal is the most significant global energy resource; over 40% of global elec-
tricity is produced from coal [1]. It is inexpensive, easy to transport and store, and a reliable
fuel for power generation, making it appealing. Similarly, in the United States, coal is an
important component of domestic power generation, accounting for 33% of electricity pro-
duction [1].
In recent years, coal prices have dropped significantly, which can be attributed to
environmental requirements on emissions, emphasis on incorporating renewable resources
into generation portfolios, and technological advances that have lowered the cost of cleaner-
burning natural gas power plants. As such, the global coal market has seen drastic changes,
and utilities in the U.S. have begun to reconsider their fuel planning strategies.
Our work is motivated by these fundamental changes to coal as a commodity, and
the impacts these changes have on global markets as well as domestic energy consumers. We
first seek to understand how price declines affect the international movement of coal from
countries that are net exporters to those that are net importers. We then shift our focus
and determine how a domestic electrical utility and its customers can economically benefit
from this changing environment.
This dissertation is comprised of three chapters that study the changes resulting
from recent coal price developments and provide the groundwork for a consumers to adjust
their market position as commodity prices evolve. We construct formulations that model
international shipping patterns and determine purchase decisions for a utility. In doing so,
we develop a general method for incorporating risk measures in forward decisions under price
uncertainty.
1
|
Colorado School of Mines
|
In Chapter 2, we present a model that minimizes the cost of shipping global thermal
coal. We develop a deterministic, mixed integer program that optimizes movement from
net exporting counties to net importing countries. We account for the volume and quality
specificationsrequiredbyimportersaswellasthephysicalconstraintsassociatedwithimport
and export ports and vessels transporting the resource. Our input data are deterministic
and include coal prices that are defined by product quality and shipping route costs, which
are based on vessel size. We solve for the minimum cost solution for serving global demand
and compare our results to observed coal flows. Using our model, we conduct case studies
to understand the effects of sub-optimal policies, such as inefficient government mandates
on the source from which China receives its coal, and the impacts of the Panama Canal
expansion on global shipping patterns. This model provides a framework that can be used
to more completely understand the global coal market.
In Chapter 3, we develop a methodology to determine an optimal forward purchase
strategy for a coal consumer under price uncertainty. This includes a price regression model,
a facility location model, and a multi-stage stochastic program. The price regression model
consists of a linear trend to capture the downward movement of coal prices and their depen-
dencyonnaturalgasprices, aperiodictrendtomodelseasonality, andavectorautoregressive
lag-one model to account for autocorrelation. We generate future price realizations and use
a facility location model to reduce the number of scenarios to keep our optimization model
tractable, while maintaining the statistical properties of the sampled prices. We formulate
a multi-stage stochastic program that minimizes the expected cost of coal procurement and
the conditional value-at-risk (CVaR). We develop a general and notationally compact nested
formulation that is time consistent and implement our model to determine optimal coal pur-
chase strategy under price uncertainty. We compare the results to those obtained by solving
an expected CVaR (E-CVaR) model [2], which is an accepted time consistent risk measure
in a multi-stage environment and show that E-CVaR may be better model for minimizing
risk in a multi-stage setting. Additionally, we develop an out-of-sample testing methodology
2
|
Colorado School of Mines
|
that allows us to determine the cost of our multi-stage program solution under new price
realizations to assess model performance.
In Chapter 4, we apply the E-CVaR model to assess the coal procurement strategy
for a utility in the United States (to which we refer as the Utility) that has the option
to engage in forward purchases with delivery scheduled for future years. The Utility faces
similar physical requirements those discussed in Chapter 1, including supply and demand
obligations and coal quality specifications. Currently, the Utility has an accepted purchase
plan that outlines a strategy for hedging against future coal price with forward purchases.
Using the E-CVaR model, we show that such a plan is not optimal given the forecast we
produce using our regression model, and that the Utility can significantly decrease both its
expected coal procurement cost and CVaR by reducing its volume of forward purchases. We
anticipate that the Utility will use our solution when purchasing coal in April, 2017.
Lastly, in Chapter 5, we summarize our contributions and suggest ideas for future
work.
3
|
Colorado School of Mines
|
CHAPTER 2
OPTIMIZING GLOBAL THERMAL COAL SHIPMENTS
This paper has been submitted to Omega. We have received referee responses, and have
revised and resubmitted the paper on November 26, 2016. It is coauthored by Dr.
Alexandra Newman, Dr. Cameron Turner, and Casey Kaptur of RungePincockMinarco.
2.1 Abstract
Thermal coal is used to produce energy; with changing emissions standards and advents
inrenewabletechnology, thethermalcoalmarkethasseensignificanttransformationoverthe
past decade. We develop a mixed-integer optimization problem that seeks to minimize ship-
ment costs while meeting demand for thermal coal, and which respects quality constraints,
supply limits, and port capacity; we use this model to analyze the following scenarios: (i)
a counterfactual setting in which we compare historical shipping patterns to model results
using a 2012 base year; (ii) the explicit effect of Chinese mandates on coal shipments; (iii) the
impact on our shipping patterns of reduced Chinese and Indian demand; (iv) the effects of
the Baltic Dry Index and oil prices; and (v) a comparison of shipments prior and subsequent
to Panama Canal expansion. Our work can be used to inform policy, study responses to vari-
able price and demand scenarios, and provide insight to both coal producers and consumers
about the international coal market. For example, removal of mandates set by the Chinese
government to fill its own demand decreases coal flows from Northern to Southern China by
56%, which has a spill-over effect on European and American markets; and, expansion of
the Panama Canal leads to only modest shipping increases through the canal (6.7%), with
more coal originating from Colombia serving Asian demand.
4
|
Colorado School of Mines
|
2.2 Introduction and Background
Coal is mined for two main purposes: energy production and steel making. Thermal
coal is lower in both quality and price and is used in energy production. Metallurgical (or
coking) coal is higher in quality and price, and is used in steel making. This paper focuses
on the transportation of thermal coal; over 6.3 billion tonnes were burned in 2010, which
represents a 70% increase from consumption levels of 3.7 billion tonnes in 2000 [3]. As
of 2013, coal accounted for 41% of global electricity production [4]. Despite the recession
in 2009, global thermal coal demand levels have resisted abatement. In fact, consumption
has increased annually for the past decade [5]. As a resource, thermal coal is cheap and
dependable; extraction and transportation procedures are well-established and economically
viable. According to the International Energy Agency (IEA), electricity will continue to be
the fastest growing component of the global energy mix over the next two decades and coal
is forecast to remain the predominant fuel for electricity generation ([1], [5]). IEA predicts
thermal coal production growth to range between 0.08% per year to 1.75% per year (56
million (MM) to 132 MM tonnes per year) over the 2012-2035 period, and expects thermal
coal consumption to range between 7.4 billion tonnes per year and 9.2 billion tonnes per year
by 2035 [1].
Most increases in thermal coal consumption are driven by significant demand growth
over the past decade in developing countries such as China and India. In 2011, coal produced
47% of energy in Organization of Economic Cooperation and Development countries, while
80% of China’s energy and 74% of India’s energy was supplied by thermal coal. Though
recent thermal coal consumption forecasts tend to be less bullish about growth in consump-
tion (the EIA reports that Chinese coal consumption growth fell to 1% in 2013 and was
essentially flat in 2014 [6]), both IEA and BP continue to forecast significant coal demand
globally. Even less aggressive forecasts predict continued growth in Chinese coal consump-
tion through the end of the decade before leveling off [7]. In fact, data released by the
Chinese government at the end of 2015 showed that previous consumption levels and growth
5
|
Colorado School of Mines
|
forecasts had been underestimated by about 11% [8].
Demand for thermal coal in countries that are members of the Organization of
Economic Cooperation and Development has remained relatively stagnant or has declined
slightly over the past decade due in large part to emissions regulations and a push for al-
ternative energy sources (which are better hedged with more flexible and responsive natural
gas generators than coal generators). Additionally, technological advances in fracking in
North America have made natural gas costs competitive with and, in many cases, cheaper
than, coal, which has encouraged consumers to displace coal-fired resources with cleaner
burning gas. Demand reduction in Organization of Economic Cooperation and Development
countries, coupled with demand growth in countries that are both net importers and large
coal consumers (such as China and India), highlight the importance of an international coal
market; net importers must rely on this growing global market to meet their needs [3]. In
fact, growth rates in inter-regional thermal coal trade are expected to be higher than growth
rates in consumption; the IEA predicts thermal trade to grow by between 1.4% and 2.7%
per annum, reaching between 1,320 MM and 1,827 MM tonnes by 2035 (up from 960 MM
tonnes in 2012) [1].
TheglobalmarkethastypicallybeendividedintotheAtlanticandthePacificspheres.
Historically, the Pacific market consisted of Japanese, Korean, Chinese, and Indian demand,
supplied by Australian and Indonesian coal, while the Atlantic market consisted of Ameri-
can and European demand, supplied by North and South American, European, and South
African coal. However, with increasing demand from China and India and stagnant or de-
creasing demand in North America and Europe, the distinction between these two markets
has blurred as opportunities arise to supply coal to the growing Asian markets. As these two
spheres blend into a single, global marketplace, coal is undergoing commoditization; its price
is less defined by where it comes from and increasingly by its quality measurements (such as
heat content, ash content, and sulfur content). Consequently, considering a minimum-cost
solution to coal demand differentiated by quality measurements is a natural evolution of the
6
|
Colorado School of Mines
|
market ([9], [3]).
In general, the global thermal coal trade is a complex and volatile business. These
complexities include:
1. Coal supply specifications: coal is not a homogeneous substance. Different coals from
different regions have different coal qualities, including varying heat (energy), ash, and
sulfur contents.
2. Coal demand specifications: each coal consumer has its own specifications, defined by
local government regulations or generator capabilities.
3. Blending capability: some coal consumers can blend coals with different qualities to
meet specifications. Others cannot or choose not to blend.
4. Coal pricing: as a result of different coal specifications and regional market conditions,
different coals have different prices.
5. Export terminal capacities and other parameters: each export terminal has finite ca-
pacity as well as draft restrictions that may preclude receiving vessels of certain sizes.
6. Import terminal capacities and other parameters: each import terminal has parallel
restrictions regarding, e.g., capacity, draft, and vessel compatibility.
7. Export terminal product mix: many coal terminals handle both thermal and metal-
lurgical coal. Higher value metallurgical coal generally receives priority from these
terminals, thereby restricting thermal coal capacity. This is particularly important in
western Canada, Queensland (Australia), the US East Coast, and the US Gulf Coast.
8. Coal availability: only certain coals with their own specifications and availabilities can
be shipped from a particular export terminal. Moreover, the availability of these coals
may not equal the terminal capacity.
7
|
Colorado School of Mines
|
9. Vessel types and rates: a variety of different types of bulk carriers are used in the coal
trade. Vessel rates vary by ship category and distance traveled.
10. Price volatility: coal prices and vessel costs vary significantly over time.
This research develops a minimum cost solution to supplying global thermal coal
demand. Our formulation assumes that all inputs are known with certainty, and includes
coal availability, quality requirements, prices, and shipping costs. We allow for blending
differenttypesofcoalatdemandnodestoachieverequiredcoalqualities, andareconstrained
by port capacities and total supply of different coal types. We compare our cost-minimizing,
international coal shipment plan against that from an historic year (2012). We can neither
turn back time nor can we act as a global decision maker, enforcing our shipment plan
along the trade routes included in our model. However, our work has implications for policy
makers as well as coal suppliers and purchasers in that it can be used to study macro-
scale, international coal markets and to assess trade agreements, shipping legislation and
environmental regulation, inter alia, and the effects these have on coal trade. Additionally,
examining our optimal shipment plans can aid in economic decision making: investment
banks have expressed interest in using this tool to evaluate financial decisions associated
with port expansion and mining operations. Coal suppliers find this work useful in assessing
where their likely clients reside in order to best focus their resources; results from this
model have been applied to confirm producer marketing efforts. Similarly, coal consumers
can leverage this research to determine their best purchase strategy based on international
market conditions.
This research develops a minimum cost solution to supplying global thermal coal
demand. Our formulation assumes that all inputs are known with certainty, and includes
coal availability, quality requirements, prices, and shipping costs. We acknowledge that
this simplification does not capture uncertainty implicit in model inputs, and point the
interested reader to an application of stochastic programming to coal procurement under
price uncertainty for a United States utility [10]. The implementation of such a technique
8
|
Colorado School of Mines
|
requires significant data collection and modeling efforts, and is beyond the scope and interest
of our industry partner (RungePincockMinarco). Furthermore, as shown in [10], not all
applications benefit from the conclusions provided by a stochastic program. In this paper,
we restrict ourselves to the examination of various scenarios, e.g., the effects of high and low
Baltic Dry Index and oil prices, as well as the impacts of high and low Chinese demand (see
Section 2.6).
We allow for blending different types of coal at demand nodes to achieve required coal
qualities, and are constrained by port capacities and total supply of different coal types. We
compare our cost-minimizing, international coal shipment plan against that from an historic
year (2012). We can neither turn back time nor can we act as a global decision maker,
enforcing our shipment plan along the trade routes included in our model. However, our
work has implications for policy makers as well as coal suppliers and purchasers in that it
can be used to study macro-scale, international coal markets and to assess trade agreements,
shipping legislation and environmental regulation, inter alia, and the effects these have on
coal trade. Additionally, examining our optimal shipment plans can aid in economic decision
making: investment banks have expressed interest in using this tool to evaluate financial
decisions associated with port expansion and mining operations. Coal suppliers find this
work useful in assessing where their likely clients reside in order to best focus their resources;
results from this model have been applied to confirm producer marketing efforts. Similarly,
coal consumers can leverage this research to determine their best purchase strategy based
on international market conditions.
The remainder of this paper is organized as follows: Section 2.3 provides a litera-
ture review with a focus on transportation networks with blending. Section 2.4 describes
the data collected to populate the parameters; Section 2.5 follows with a discussion of our
mathematical formulation. Section 2.6 provides several case studies, after which Section 2.7
concludes.
9
|
Colorado School of Mines
|
2.3 Literature Review
Studying the global thermal coal market is not new; the International Energy Agency
has conducted large-scale supply-demand balancing (including future supply and demand
forecasting)[3], tantamounttosolvinganaccountingproblem. Additionalworkfocusesmore
heavily on producer decisions regarding whether to sell locally or to export, and on annual
decisions regarding whether or not to invest in mining technologies, making simplifying
assumptions about the nature of importer activity and neglecting import and export port
capacity constraints [11].
In a seminal paper, Samuelson [12] discusses the application of linear programming
to the economic equilibrium problem introduced by Enke [13] in which geographically sepa-
rate markets have their own domestic supply and demand curves and between which trans-
portation of products is allowed. The resulting solution determines which markets are net
importers and exporters, how much product is shipped to serve demand, and prices at equi-
librium. We incorporate more detail through the use of integer variables that allow us to
include the fixed cost of ships transporting coal, as well as to bound blending capabilities
at demand nodes; we address the complications that arise if we relax integrality in detail in
Section 2.6.
Gabriel et al. [14] address bi-level models for hierarchical decision making and equi-
librium models for decision-maker interaction. We assume that all agents are price-takers,
i.e., that costs are not influenced by trade flows, that all agents are symmetric in their abil-
ity to produce and consume coal, and that no single agent is sufficiently large to influence
market conditions, e.g., prices based on global supply and demand. We also assume that
supplies and demands are given for our year-long timeframe. That is, we are interested in
efficient coal shipping patterns for a set of static conditions, rather than in the evolution of
the market with agents sufficiently influential to change supplies, demands, and prices that
other agents face.
10
|
Colorado School of Mines
|
Papageorgiou et al. [15] minimize the fixed and variable costs to ship raw materials
that are blended into homogenous final products. The problem of moving and blending
coal has been studied on a smaller scale (Lai and Chen [16], Shih [17] and Liu and Sherali
[18]); however, these authors focus on a single utility in Taiwan while our research attempts
to understand the global thermal coal market. Problems with similar considerations are
relevant to commodities other than coal; for example, Bilgen and Ozkarahan [19] solve a
mixed-integer program to determine optimal blending and shipping of grain across multiple
time periods.
Some network design optimization models develop both routes and schedules for
container ships; our model assumes routes are predetermined and instead chooses how to
ship coal to meet demand and specification requirements. Work on routing and scheduling
over the past two decades is summarized in survey papers by Christiansen et al. ([20], [21]).
Mudrageda et al. build an economic-equilibrium model to inform business strategies for a
petroleum shipping company operating in the Eastern United States marine transportation
market [22]. Perakis and Papadakis minimize the cost to operate a fleet that must transport
a given capacity of cargo over a trade route in a specified amount of time by varying the
speeds at which fully loaded and unloaded ships travel ([23], [24]); they consider both the
fixed cost to operate vessels and the variable cost incurred per ton of cargo, and note that
when fleet composition is allowed to change, fixed costs have a larger impact on the optimal
solution. In our work, we allow selection of different-sized vessels over shipping routes, so
we must include fixed costs in our objective.
Our research combines elements from much of this past work; we use available data on
global coal markets to populate our parameters, and formulate a global cost-minimization
model with underlying network structure which focuses on consumer decisions regarding
sourcing coal to satisfy demand and quality constraints by allowing for blending of multiple
coal types at demand nodes. We do not take into account domestic activity or producer
investment decisions, and we predefine available routes based on port and passage capacities
11
|
Colorado School of Mines
|
as well as based on regional coal qualities.
2.4 Data
Our research combines elements from much of this past work; we use available data on
global coal markets to populate our parameters, and formulate a global cost-minimization
model with underlying network structure which focuses on consumer decisions regarding
sourcing coal to satisfy demand and quality constraints by allowing for blending of multiple
coal types at demand nodes. We do not take into account domestic activity or producer
investment decisions, and we predefine available routes based on port and passage capacities
as well as based on regional coal qualities.
Port capacities are used in the model in two ways. Viable paths from source to
sink rely on the maximum vessel size that can access both import and export ports, which
we define as the smaller of the two largest vessel capacities that can be accommodated at
each port. Data on the largest ship size that can access a port originate from publicly
available information at the World Port Source website [25], which provides channel width
and anchorage depth.
The second relevant measure of port capacity is annual throughput, defined by the
numberofoperationalberthsdedicatedtoshippingorreceivingcoal, aswellastheamountof
thermal versus metallurgical coal a port either ships or receives. In some cases, port through-
put is given by external constraints, such as rail operations. For example, the Richards Bay
Terminal in the Republic of South Africa has a nominal capacity of about 90 MM tonnes
per year, but the South African National Railroad, Transnet, struggles to deliver more than
75 MM tonnes per year. News sources such as Platts Coal Trader [26], Argus Coal Daily
International [27], and Coal Export Services International [28] provide data regarding how
much coal has historically flowed through a port in a given year. Our model includes about
180 import coal facilities with a total capacity of 1.5 billion tonnes per year, and 75 coal
export terminals (or generic export regions as in the case of Indonesia where off-shore barge
terminals make it difficult to determine specific port data), totaling an export capacity of
12
|
Colorado School of Mines
|
2.0 billion tonnes per year.
Coal specifications define the price of a specific type of coal, as well as dictate whether
a consumer is able to accept and burn a coal. This research focuses specifically on heat
content (measured in MMBTU, or million BTU, per pound), sulfur, and ash content (both
measured as a percent of total weight). Coal is a commodity that has characteristics that
depend on where the coal is mined. For example, coal from Appalachia tends to have a
high heat content which burns more efficiently (i.e., more energy is released per unit of
coal burned). However, some Appalachian coal is high in ash and sulfur content, which
is less attractive from an environmental perspective. Coal from the Powder River Basin
in Wyoming has lower heat, sulfur, and ash content than Appalachian coal, which makes
it less efficient but cleaner to burn. Data on regional coal quality stems from multiple
sources, including Platts [26] and Argus [27] [29], which define coal prices. In the United
States, sources such as the Burlington Northern Santa Fe Railroad [30] and the Union Pacific
Railroad [31] provide detailed specifications of coal produced at mines served by railroads. In
ourmodel, wedefine39availablecoaltypesbasedonregionalspecificationsandavailabilities.
For example, we account for two coal types originating in the Powder River Basin, one in
the Illinois Basin, and two in Appalachia. All coal types have predefined specifications. We
also define which export ports are able to supply which coal types based on deliverability.
For example, Powder River Basin coal can ship from both West and Gulf Coast ports while
Appalachian coal is available from East and Gulf Coast ports. More than one port can
export a given coal type.
Cost data include coal prices (based on coal quality), freight rates for shipping, and
tariffs associated with using canals. Coal prices are a product of a proprietary calculation
in which data from various sources are combined and are determined at each export facility
(implying that the same coal type may have different prices based on the export facility
from which it is shipped). Freight rates are calculated using the Baltic Dry Index, which is
a measure of the cost to ship dry goods by sea, and are specific to ship type, oil prices, and
13
|
Colorado School of Mines
|
route distance. A single, shortest route is calculated between every export port and import
port for each ship type using routing tools such as the World Port Source [25] and a global
ports database [32]. Tariff rates for the Panama Canal [33], Suez Canal, and Turkish Straits
are determined for different ship sizes. These data are combined to determine a price per
ship sailing from each export port to each import port carrying a single type of coal.
In addition to the above sources, data is verified and supplemented by industry
experience, client insight, and proprietary information.
2.5 Model Formulation
Our global thermal coal optimization model minimizes transportation costs, and
possesses an underlying network structure with supply and demand nodes (where coal enters
andexitsthemodel,respectively),transshipmentnodes(ourports),andarcs(thepermissible
paths over which coal can be shipped). The model hass supply and demand capacities, arc
costs (defined to be the cost to procure and ship coal from supply to demand node), and
arc capacities (which are defined by the capacities of the ships traveling along a path from
source to sink). Our formulation relies on the construction of permissible shipping paths,
which are defined as quintuplet sets of export port, import port, demand node, ship size,
and coal type. The existence of these permissible paths is dependent on maximum vessel
size accommodated at import and export ports and coal types available at supply nodes.
We employ positive, continuous variables, x , that represent the tonnes of coal type θ
eidθs
that are shipped from export port e to demand node d through import port i on ship size s.
Network model construction, properties, and algorithms are discussed in detail in standard
operations research texts, including Rardin [34] and Bazaraa et al. [35].
However, our model violates a pure network structure because our arc capacities
are functions of a discrete number of ships traversing a path; if no ships are sent from a
supply node to a demand node (which is a model decision), that arc has a capacity of 0.
We use positive, integer variables, y , which determine the number of ships traveling
eidθs
along permissible shipping paths. Sending under-filled ships is discouraged because of the
14
|
Colorado School of Mines
|
corresponding fixed charge, which we include because, in our application, spot charters can
be used to ship coal, especially in Japanese and Korean markets. That is, we cannot assume
long-term shipping contracts, which might obviate the need for such fixed costs via a “cost
smoothing” argument, nor can we claim that fractional ships are to be interpreted as load
that is consumed over multiple time periods; doing so ignores potential holding and re-
handling costs incurred from one time period to the next, inventory capacity constraints at
demand nodes, as well as coal quality degradation, which may result in violation of our coal
quality requirements. Simple rounding heuristics to construct an integer solution from the
linear programming relaxation prove to be elusive (see Section 2.6). Similar simplifications
regarding the elimination of blending coal types are also invalid; our model captures the flow
of different quality coals from various sources, a fundamental aspect of the international coal
market.
In a pure network model, units of flow are not distinguished from one another. Multi-
commodity flow models allow distinct type of units to be maintained on each arc. Our model
resembles a multi-commodity flow model in that we track different coal types flowing over
each arc; we deviate from a traditional multi-commodity flow model in that we allow blended
coal types to serve a single demand at each node in a way that meets coal specification
requirements. We use binary variables, z , to ensure that at most a pre-defined maximum
dθ
number of coal types, u , which is defined by individual capabilities at coal-fired plants,
d
are blended at each demand node (a construct whose inclusion is dictated by real-world
practice). The sources from which these coal types are received and the type of ships which
carry the coal from source to demand node are not restricted as long as the number of coal
types used to fill demand is not violated. To this end, the binary variables z used to limit
coal types are indexed only by demand node and coal type.
Figure 2.1 and Figure 2.2 illustrate viable flows. The example contains two export
ports (e = {1,2}), two import ports (i = {1,2}), two demand nodes (d = {1,2}), and two
coal types (θ = {1,2}). In Figure 2.1, demand node 1 is receiving coal type 1 and coal type
15
|
Colorado School of Mines
|
2 from export port 1, while demand node 2 is receiving coal type 1 from both export ports
1 and 2. Though different sources are supplying coal, demand node 2 is only receiving a
one coal type, while demand node 1 is receiving two types of coal from a single source (on
multiple ships). In this example, both demand nodes are currently receiving the maximum
number of coal types which they are capable of blending.
Demand node 2 can increase the amount of coal type 1 that it is receiving from any
source because it is already consuming that coal type. However, in order to additionally
accept coal type 2, z would also have to assume value 1, which it cannot because the
22
maximum coal types constraint is already binding. Demand node 1 can accept either coal
type 1 or 2 from either source as it is already receiving both types. As such, in the second
solution (Figure 2.2), demand node 1 can additionally source coal type 1 from export port
2 without violating the maximum coal types constraint.
We now state the notation and mathematical expressions for our optimization model,
which we solve for a one-year time horizon; correspondingly, supplies, demands, and capac-
ities are defined on an annual basis.
Sets
• s ∈ S: Set of all ship types; s ∈ {c = capesize,p = panamax,h = handysize}
• e ∈ E: Set of all export ports
• i ∈ I: Set of all import ports
• d ∈ D: Set of all demand nodes
• θ ∈ Θ: Set of all coal types
• sˆ∈ Sˆ : Set of all ship types that can leave export port e
e
• sˆ∈ Sˆ : Set of all ship types that can enter import port i
i
17
|
Colorado School of Mines
|
• dˆ ∈ Dˆ : Set of all demand nodes accessed through import port i
i
• θˆ ∈ Θˆ : Set of all coal types that can be exported from export port e
e
• P: Set of all eligible paths, where each path is given as an export port, import port,
demand node, coal type, ship type quintuplet:
(e,i,dˆ,θˆ,sˆ)
Parameters
• δ : Demand at node d (tonnes)
d
• σ : Supply of coal type θ (tonnes)
θ
• γ : Capacity of ship type s (tonnes)
s
• v : Variable cost of coal type θ traveling to demand node d from export port e
eidθs
through import port i on ship type s ($/tonne)
• c : Fixed cost per ship type s carrying coal type θ to demand node d from export
eidθs
port e through import port i ($/ship)
• κˆ : Capacity of export port e (tonnes)
e
• κ˜ : Capacity of import port i (tonnes)
i
• h : Heat content of coal type θ (MMBtu/lb)
θ
• f : Sulfur content of coal type θ (% sulfur)
θ
• a : Ash content of coal type θ (% ash)
θ
• h : Minimum heat content requirement for demand node d (MMBtu/lb)
d
• a¯ : Maximum sulfur content requirement for demand node d (% ash)
d
• f¯ : Maximum ash content requirement for demand node d (% sulfur)
d
18
|
Colorado School of Mines
|
• u : Maximum number of sources from which demand node d can receive coal
d
• M A sufficiently large number
Variables
• x : Amount of coal type θ from export port e through import port i on ship type
eidθs
s serving demand at node d (tonnes)
• y : Number of ships of type s carrying coal type θ traveling from export port e
eidθs
to import port i to serve demand at node d (ships)
1 if coal type θ serves demand at node d
• z :
dθ
0 otherwise
2.5.1 Objective
min c y + v x
eidθs eidθs eidθs eidθs
(e,i,d,θ,s)∈P (e,i,d,θ,s)∈P
X X
The objective sums the product of the fixed cost per ship journey (specific to ship type,
coal type, and shipping path) and the integer number of ships designated on each path
with the product of the variable cost per tonne of coal (specific to ship type, coal type, and
shipping path) and the tonnes of coal shipped along that path.
2.5.2 Constraints
x ≤ σ ∀θ ∈ Θ (2.1)
eidθs θ
e,i,d,s:
X
(e,i,d,θ,s)∈P
Constraint (2.1) ensures that the sum over all paths on which coal shipments occur must
remain at or below supply limits for every coal type.
x ≤ κˆ ∀e ∈ E (2.2)
eidθs e
i,d,θ,s:
X
(e,i,d,θ,s)∈P
19
|
Colorado School of Mines
|
of coal shipped over the path are forced to assume a value of zero.
y ≤ Mz ∀d ∈ D,θ ∈ Θ (2.9)
eidθs dθ
e,i,s:
X
(e,i,d,θ,s)∈P
Constraint (2.9) only allows a variable z to assume a value of 1 if the coal type by which
dθ
it is indexed is either: (i) already being used at the demand node by which it is also indexed
or if (ii) the maximum number of coal types used at the demand node is not yet reached
(this is controlled by constraint (2.10)). If a z variable assumes a value of 1, then the y
dθ eidθs
indexed by the same coal type and demand node can assume a positive value, i.e., we can
send ships carrying coal type θ to demand node d.
z ≤ u ∀d ∈ D (2.10)
dθ d
θ∈Θ
X
Constraint (2.10) ensures that a single demand node cannot receive its coal from more
than a maximum number of sources, regardless of demand node.
x ≥ 0 ∀e,i,d,θ,s
eidθs
y ≥ 0, integer ∀e,i,d,θ,s
eidθs
z ≥ 0, binary ∀d,θ
dθ
All x variables are non-negative; all y variables are non-negative and integer; and all
z variables are binary.
2.6 Results
WecodeourformulationinAMPLversion20140908andruninstancesinCPLEX12.6.0.1
on a Dell PowerEdge R410 with 12 GB of RAM, a 160 GB hard drive, and 16 processors
with 2.72 GHz each. Model instances contain more than 133,000 variables and 70,000 con-
straints, many of which can be removed with presolve, and the result of which is a constraint
set primarily dominated by a network structure. Average solution time is 12 seconds to
obtain a 0.15% optimality gap, obviating the need for special decomposition procedures or
other means to expedite solutions. For larger scenarios, this network structure could be ex-
21
|
Colorado School of Mines
|
ploited by solving the linear programming relaxation and developing sophisticated rounding
heuristics for the ship movement (y) and coal sourcing (z) variables. However, heuristics
can be difficult to construct [36], and this has been our experience here. Specifically, for
the base case (i), presented directly below, we solve our model with integrality requirements
relaxed for our ship movement variables, y. We then (i) round each variable up or down to
the nearest integer or (ii) round each variable up to the next whole number. With neither
scheme when we resolve our model with this rounded solution fixed (and optimize over all
other variables) do we obtain a feasible solution. (Generally, there exist violations of demand
and/or port capacity.)
We present several case studies: (i) a counterfactual scenario in which we compare
historical shipping patterns to model results using a 2012 base year; (ii) the explicit effect
of Chinese mandates on coal shipments; (iii) the impact on our shipping patterns of reduced
ChineseandIndian demand; (iv)theeffectofvariation intheBalticDryIndexand oil prices;
and (v) a comparison of shipments prior and subsequent to Panama Canal expansion. We
do not include specific costs or other sensitive data to maintain the confidentiality required
by our industry sponsor, RungePincockMinarco.
2.6.1 Historical shipping patterns using 2012 as the base year
We define higher quality coal to be lower in ash and sulfur content and higher in heat
content, and, therefore, more expensive. We highlight differences between our model output
(Figure 2.4) and actual 2012 coal flows (Figure 2.3) from 2012 (using IEA data). The width
of each path is proportional to the amount of coal flowing from source to sink. Figure 2.3
shows more coal moving from Northern to Southern China in accordance with government
requirements favoring Chinese coal consumption; Figure 2.4 depicts the case in which we do
notforceChinesedemandnodestopurchaseChinesecoal. Hence, ititismoreeconomicalfor
Chinese demand to be served with additional Indonesian and Australian coal. The model
also offsets lower quality Indonesian coal in India with higher quality South African coal
whose displacement from Europe forces more European demand to be satisfied with coal
22
|
Colorado School of Mines
|
by incorporating constraint (2.11) into our optimization model:
x ≥ c δ ∀θ ∈ Θ (2.11)
eidcθs dc
ecX,i,dc,s: Xdc
(ec,i,dc,θ,s)∈P
where e and d are Chinese export port and demand nodes, respectively. The parameter c is
c c
the percentage of Chinese demand that must be met by Chinese coal, and the product on the
right-hand side is equal to difference between total Chinese-reported coal consumption and
net imports to reflect historical government mandates; we set c accordingly. Running our
optimization model with and without this constraint allows us to understand the impacts of
requiring Chinese demand centers to purchase coal from Chinese producers.
Figure 2.7 and Figure 2.8 show these impacts. For the case in which Chinese demand
is required to be filled by Chinese coal, Chinese sources account for almost half of the coal
consumed in China. Removal of this requirement produces a solution that demonstrates
that it is much more economical to purchase coal from Australia and Indonesia; Chinese
coal serves less than a quarter of Chinese demand. This reduction has a cascading effect on
global shipping patterns.
Figure 2.7: 2012 Chinese coal sources with government mandates in place. Almost half of
Chinese demand is served by Chinese sources.
As coal from Indonesia and Australia is sent to China, its displacement from other
demand nodes must be remedied. Our model still shows Australian coal serving Japanese
demand, but we see Japan also sourcing from Indonesia, Russia, Alaska, Canada, and even
Europe. (The higher quality coals offset the cheaper, lower quality coal from Indonesia.)
25
|
Colorado School of Mines
|
Figure 2.8: 2012 Chinese coal sources when the model may source coal most economically.
Less than a quarter of Chinese demand is served by Chinese sources.
South Korea and Malaysia still source coal from both Australia and Indonesia, but India is
much less dependent on Indonesian coal in our model, obtaining a higher volume from South
Africa. In turn, the displacement of South African coal from Europe means that our model
serves European demand with North and South American coal, and we see larger volumes
shipped across the Atlantic.
2.6.3 Impact of reduced Chinese and Indian demand
In recent years, China, committed to reducing overall emissions, has worked on in-
creasing its renewable resource portfolio, and has also faced an economic slow-down. Ad-
ditionally, the country has improved its infrastructure, allowing Northern Chinese coal to
serve Southern Chinese demand at economically viable prices, providing us with an example
in which land transportation influences shipments. As such, there has been a decrease in net
imports to China, with volumes falling from around 250M tonnes in 2012 to 200M tonnes in
2015 [37]. As a result, coal exports in Australia and Indonesia have decreased in both price
and export volumes. This is partially due to the fact that China has been a large consumer
of both Australian and Indonesian coal. Figure 2.9 shows the decrease in overall Australian
exports as Chinese volumes are reduced.
While Figure 2.9 shows a major decrease in Australian exports, Indonesian exports
remain stable, largely due to Indian demand and loose emissions requirements in both India
and China, which allow both countries to continue to burn relatively cheap and low quality
26
|
Colorado School of Mines
|
Figure 2.9: Australian and Indonesian total exports under falling Chinese demand.
Indonesian coal.
2.6.4 Impact of Baltic Dry Index and oil prices
The freight rates that our model uses to establish the fixed cost of ships depend both
on the Baltic Dry Index (BDI) and on oil prices. Because we use 2012 as a base year, the
BDI and oil prices we include are higher than their present values, and, in fact, have seen
a significant amount of volatility over the past five years. Figure 2.10 shows the results of
three model runs: (i) a base case with 2012 observed BDI and oil prices; (ii) a case in which
we increase BDI and oil prices by about 50% to reflect high 2011 prices; and (iii) a case in
which we decrease BDI and oil prices to today’s levels.
For simplicity, Figure 2.10 only includes the highest importers and exporters. Varying
BDI and oil prices results in small fluctuations in total imported tonnes, stemming from an
importer receiving coal from a source that is not included as a top exporter, and resulting
from a reallocation of sources from which coal is received. This shifting of sources is not
so apparent in Chinese consumers as coal is procured from nearby sources such as China,
27
|
Colorado School of Mines
|
Indonesia, and Australia, and so is not significantly impacted by changing freight rates. We
see large swings in sources from which Europe and India receive coal. Most notably, when
BDI and oil prices decrease (resulting in cheaper shipping), Europe tends to source more
coal from Colombia, and India leans more heavily on South African coal. When BDI and
oil prices increase, India imports more Indonesian coal as it is cheaper to ship. In this case,
Europe decreases the coal it receives from Colombia and increases the volumes it imports
from South Africa. Similarly, though we do not see large changes in Chinese sources, the
rest of Asia shifts to import coal from the United States instead of Colombia, as the United
States West Coast is closer, resulting in reduced transportation costs. This same pattern
holds for the United States; rather than importing Colombian coal in the high BDI case, the
US instead sources coal domestically to save on shipping costs.
2.6.5 Panama Canal prior and subsequent to its expansion
We compare model output before and after Panama Canal expansion. Prior to ex-
pansion, the largest vessels that could traverse the Panama Canal were panamax ships, with
a capacity of 70,000 tonnes. Following expansion of the canal, the post-panamax vessel with
nominal capacity of 130,000 tonnes can now pass through. Atlantic Basin coal refers to
coal sourced from countries on the shores of the Atlantic Ocean (including the Caribbean
and Baltic Seas). This includes the US East Coast and Gulf Coast, Europe, Colombia
and Venezuela, the Baltic (Poland, Ukraine, western Russia) and the Canadian East Coast.
Pacific Basin coal is generated by countries in the Pacific Ocean and includes the US and
Canadian West Coasts, Australia, Indonesia, China, and eastern Russia. Analysts have sug-
gested that the canal expansion, combined with presumably lower freight rates for the larger
post-panamax vessels, will facilitate a substantial penetration of Atlantic Basin coal into
Pacific Basin markets previously served only by coal originating from the Pacific Basin.
We relax the maximum vessel capacity which can traverse the Panama Canal from
70,000 tonnes to 130,000 tonnes. We develop freight rates for these new vessels by interpo-
lating between the rates for the traditional panamax class vessels and the rates for larger
29
|
Colorado School of Mines
|
capesize vessels. Further, we change the fee structure for the canal, with larger vessels paying
higher tolls. In order to isolate the impact of canal expansion, we make no other changes
to the model, e.g., supply and demand volumes, port capabilities and capacities, coal avail-
ability, and coal pricing. Under these circumstances, this analysis is a zero-sum game: any
gains by Atlantic Basin suppliers must be exactly offset with losses from traditional Pacific
Basin suppliers.
Before any expansion, the model shows total coal volume through the Panama Canal
is 18.45 MM tonnes. Once post-panamax vessels use the passage, total coal volumes through
the canal increase a relatively modest 6.7%, to 19.675 MM tonnes. The canal expansion does
not seem to lead to a significant increase in thermal coal throughput volume.
Thisrelativelymodestincreasecamouflagesmajorchangesinoriginsanddestinations.
Post expansion, exports from the US Gulf Coast and Venezuelan sources traveling through
the canal decrease and are replaced by exports from Colombia. While this outcome may
appear to indicate that the Panama Canal expansion would be adverse to the U.S. coal
export industry, the actual impact is far more muted, with U.S. Gulf exports remaining
relativelystagnant. Rather, thedisplacedU.S.coalsupplantsColombiancoalsoneast-bound
movements; becausepost-expansionColombiancoalsshippedthroughthecanalarenolonger
available to traditional northwest Europe and Mediterranean markets, those markets need
to secure their coal requirements from alternative sources, which turn out to be the U.S.
Gulf Coast ports that had lost volume through the expanded canal. Similarly, pre-expansion
Colombian volumes being delivered to South American destinations are replaced by U.S.
Gulf Coast and Venezuelan coals.
Because total demand is unchanged in the two cases, the increased shipments of
Colombian coals to the northeast Asian markets must displace coals from other origins.
These reductions come primarily from Australian, Indonesian, and U.S. sources. Colombian
origins are much closer to the canal than U.S. origins and, hence, have less sailing time and
expense. DisplacedU.S.coalisredirectedtoservedemandinEuropeandtheMediterranean.
30
|
Colorado School of Mines
|
The total impact is again more muted for Australian and Indonesian origins, as the model
reassigns these relatively competitive coals to other destinations.
2.7 Conclusions and Implications
We solve a deterministic problem for optimal coal shipping patterns such that global
demand is met while coal quality constraints are respected, coal supply is not violated, and
port capacities are upheld at minimum cost. Our model results mimic historical reality with
exceptions for uneconomical government mandates, which have spill-over effects to other
shipping patterns.
Our output can provide suggestions regarding trade route decisions for suppliers
and consumers, as well as insights concerning economic viability of port expansion and
construction projects and impacts of policies such as environmental regulation of coal-fired
power. We consider cases of imposing Chinese trade mandates, and decreasing Chinese and
Indian demand on Australian and Indonesian coal markets; we explore the effects of varying
Baltic Dry Index and oil prices on international coal trade. Finally, we conduct a case study
on the expansion of the Panama Canal, and have shown that while volumes passing through
the canal only slightly increase after the expansion, large changes in sources and destinations
of shipping routes are apparent.
Our findings are of interest to many players in the international coal market. Pol-
icy makers will see value in the economic benefit of reducing government intervention in
free market operation, and environmentalists will be able to assess the impacts of cleaner
technology, global emissions agreements, and improved domestic infrastructure. Along with
yielding a macro-level view of coal trade, our model can help both consumers and producers
understand their role in the global coal market, and focus their strategies and marketing
efforts. Finally, our model estimates the value of prospective projects, such as the Panama
Canal expansion, by demonstrating the effects these may have on optimal shipping patterns
and coal flows; these results interest project financiers and managers.
31
|
Colorado School of Mines
|
CHAPTER 3
A MULTI-STAGE STOCHASTIC PROGRAM FOR FUTURE COMMODITY
PROCUREMENT
In Chapter 2, we modeled the global thermal coal market and resulting shipping
patterns. We now shift our focus to the purchase decisions a domestic coal consumer must
make given future price uncertainty. We develop a methodology in which we use observed
price data to forecast future scenarios and select representative realizations to include in a
multi-stage stochastic program. Our multistage model minimizes the cost to procure coal
along with the conditional value-at-risk. We formulate and solve a nested model that is time
consistent, and compare our results to those found solving an expected CVaR model. This
paper will be submitted to the European Journal of Operational Research and is co-authored
by Dr. Amanda Hering and Dr. David Morton.
3.1 Abstract
Fuel prices are a significant cost to electric utilities. With recent decreases in coal
prices, a utility may seek to reevaluate its coal procurement strategy while protecting itself
against future price uncertainty. In this paper, we develop a methodology that can be used
to inform purchase decisions of physical goods for delivery both now and in the future under
price uncertainty. Specifically, we develop a time consistent multi-stage stochastic program
that minimizes both the expected cost to procure coal and the conditional value-at-risk in
order to establish a purchase plan that is robust to coal price variability. To do this, we first
construct a time series model using historical data to generate stochastic price forecasts. We
then use a facility location model to reduce the number of scenarios in each stage to improve
the tractability of our optimization model. We implement both a nested conditional value-
at-risk (CVaR) model and an expected CVaR (E-CVaR) model, analyze the results of both,
and show that the E-CVaR model may be a better choice for a multi-stage problem. Finally,
32
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.