Sunteți pe pagina 1din 23

1

Integrated Test Programs for Modern Flotation Flow Sheet Development

Peter Amelunxen (corresponding author*)


Aminpro Peru S.A.C.
Cerros de Camacho 440, Dep F-17
Santiago de Surco,
Lima, Peru
peter@aminpro.com

Roger Amelunxen,
Amelunxen Mineral Processing Ltda
41961 Ross Rd,
Brackendale, B.C., Canada

Leonardo Flores
Aminpro Chile S.p.A.
Cerro San Cristobal 9511
Santiago, Chile

Thom McCord, Peilin Lan


Freeport McMoRan Copper & Gold
333 N. Central Ave,
Phoenix, AZ, USA

Abstract

Modern flotation circuit flow sheet development draws on elements of a broad field of metallurgy and
other sciences, including hardness testing and grinding circuit modeling, process mineralogy, analytical
chemistry, mathematical modeling and simulation, statistics, cost estimation, and basic economics and
finance. This paper undertakes a practical review of a typical feasibility level flotation test program and
associated flow sheet development. Using some recent copper porphyry projects as examples, all
steps are reviewed, from sample selection through production forecasting. Common problems are
identified and discussed, and some less common ones as well. It is hoped that this paper will serve as
a practical guide for metallurgists involved in test program management and process design. It is also
hoped that some of the limitations of the current methods highlighted herein will help serve as a
compass for orienting future research efforts.

1 Background
This paper presents the authors approach to flotation circuit flow sheet development. Most of our
experience has been in the development of large copper porphyry projects in the western hemisphere
and it is with this frame of reference that we have approached the present work. Nevertheless, we
believe that many of the principles described in this paper can be applied to other kinds of projects.

The scope of the typical flotation test program has undergone a lot of change over the past 15 years.
While open cycle and locked cycle tests are still useful, they are rarely used for direct scale-up
anymore. Gone are the days when rougher retention time requirements were estimated by multiplying
the batch flotation time by a factor of 2.5, or when a bankable feasibility study could be presented in a
single two-inch ring binder.
2

There have been several key drivers for these changes. They include the increased capital cost
associated with the construction of new projects. These costs, in turn, are driven by lower head grades
and a corresponding increase in the plant capacity and infrastructure required to achieve the
economies of scale. The higher costs have reduced the tolerance for risk, and as a result project stake
holders are seeking new technology and new methods for understanding and mitigating risk.
Fortunately, new technology, in the form of improved modeling techniques, better test methods, and
faster computers has evolved to meet the demand. One of the objectives of this work is to discuss how
these new techniques can be used in a comprehensive test program.

There are a lot of factors that can influence the flotation circuit design and, by extension, the test
program. The field is so broad and deep that it is impossible to cover everything in a single paper, and
space compromises are inevitable. In most cases we have elected breadth over depth, mainly because
we feel that in our field, in-depth coverage of a given issue is more common than the broader, holistic
view that we are trying to provide here.

Lastly, we should note that we, the authors, have learned much about test program development over
the past decades, and we continue to learn with each program. Unfortunately, often new learnings are
accompanied by mistakes (although we dont like to admit it). We have tried to highlight some of these
for the reader (the learnings, not the mistakes) in the sections titled What to watch out for.

2 Objectives of a Test Program


The objectives of a test program are rarely the same from one project to the next. If we consider the
objectives of the owners team, then usually there is a primary objective followed by several secondary
objectives. For example, the primary objective can be some statement that expresses a result (highest
NPV flotation circuit design), a quality standard (often the term bankable is used, although the
definition varies), and a time frame (for large projects with multiple ore types, this can be from six
months to 18 months or more).

Usually the primary objective of a test program involves the development of a technically feasible
industrial process concept. This necessarily involves scale-up and simulation work, and, therefore, the
key deliverable is much more than just a table of laboratory test results.

The secondary objectives can be numerous and detailed. They often include:

- Produce a sample of representative concentrate for the sales and marketing people
- Produce a sample of representative tailings for designing the water reclamation system,
hydraulic transport system, tailings impoundment facility, and environmental testing
- Produce the information required to size the grinding circuit
- Produce a detailed mass and water balance for the engineering team
- Produce a cost and revenue forecast and assist the mine planning/scheduling using tradeoff
analysis between grinding throughput and flotation performance
- Produce a process design criteria
- Produce an error range and confidence level of the recovery and concentrate grades
- Satisfy any strategic constraints of the owners (environmental, water, social, infrastructure, or
other)
3

Whatever the objectives of a given test program, they should be clearly stated at the outset of the
project so that if they change in midcourse (as they often do, for good reasons), then the metallurgical
team is not held accountable for the inevitable delay in the project.

What to watch out for

Objectives that are difficult, expensive, or impractical to achieve under the budget constraints. For
example, it is very difficult and expensive to generate a molybdenite separation circuit design with
the same precision as that of the bulk sulfide circuit, due to constraints on the availability and
representativeness of the samples required.
Objectives that are driven by the timing of the engineering work. If the engineering team is waiting
for metallurgical test results, then the metallurgist has a very serious problem, as this incentivizes
the team to rush their analyses, take shortcuts, and will likely incur higher costs for both the testing
and engineering. This will also add a lot of stress on the metallurgists, engineers, accountants, the
project director, and corporate vice-presidents (and not in that order). As a general rule, we
suggest that the engineering team only be commissioned after the primary metallurgical design
data has been extracted and developed from the test program.
Information or expected levels of confidence that are not commensurate with the expenditure on
test work. For example, if gold assays are not done, gold recovery will not be forecasted (so, one
should ensure they know if gold is a payable product, before deciding whether or not to include gold
assays).
Information or expected levels of confidence that are not possible with the available information.
For example, if the metallurgist includes gold assays and produces a gold model, but the geologist
has not included gold in the geologic or block model, then that also presents a problem.
Any corporate objectives, internal constraints, strategies or strategic constraints, stage gates or
quality standards that may apply. It is a good idea to get these in writing so they can be included in
the program documentation. Even if none apply, it is recommended that the metallurgist includes
language to that effect, so that it is clear that the question has been asked.
Objectives or deliverables that are vague or ill-defined. This seems obvious, but it is a common
problem. For example, the term bankability means different things to different people. In fact,
even metallurgists dont agree on the definition of bankability. Nevertheless, if the project owner
wants a bankable document, then the metallurgist needs to ensure the program meets and
exceeds that standard, whatever it is.

3 Defining the Scope of Work


To properly define the scope of work for a flotation test program requires knowledge and experience,
and even then it is difficult to get it right. Generally, at the feasibility study level, the geometallurgical
domains or ore types have already been defined. If they have not been defined, then some kind of ore
affinity analysis or cluster analysis is suggested to reduce the geological domainsusually quite
numerousto a more manageable number of geometallurgical domains, and there are numerous
examples in the literature (Montoya, et al. 2011) (Baumgartner, et al. 2011). The cluster analysis might
include such detailed and sophisticated approaches as PCA (principle component analysis) to
determine major grouping and subsequent DTA (decision tree analysis) to refine the groupings. The
latter offers the benefit of allowing the analysis to be conducted on categorical and continuous or
discrete numerical data, thus allowing for the inclusion of all data in the reduction process.

Cluster analysis or affinity analysis generally provides significant cost reduction without compromising
the granularity of the resulting models. Note that this step requires that the metallurgist to spend some
time with the geologist to develop a deeper understanding of the ore deposit, what it contains, and how
4

it is distributed. For example, it is not enough to say that the 600 million tonne ore body contains 100
million tons of skarn when it may only represent a 5 meter layer at the rim of the deposit, as opposed
to, say, a 100 million tonne pod under 200 million tonnes of oxide ore.

Once the number of ore types has been defined, the test program can be developed. Fortunately, in
flotation there is not yet a universally accepted modeling standard. Each consultant, engineer, or lab
might have a different modeling methodology and test standard. It is necessary to know who will be
performing the analysis of the results, and what model and test standards they will be using, before the
scope or cost of a program can be properly defined. This a key point to remember, as it can often catch
everybody by surprise. For example:

It can be difficult for a junior engineer to explain to their manager why they cant use their
preferred engineer or consultant without having to repeat some or all of the test work
It is unpleasant for the process consultant or engineering company who must tactfully discount
or ignore prior work because the test methodology used in prior programs is not compatible
with their methods, procedures or database. In the worst cases, we have seen test programs
that are so poorly planned and executed that the data is misleading.
It can be discomfiting for the project director when the engineering company refuses to sign off
on the flotation circuit design criteria because they dont understand or endorse the testing or
modeling methods that were used to develop it.

It is also extremely important to know what Qa/Qc measures they, and their analytical labs, have in
place and how they are enforced. The Qa/Qc cost can be up to 10% of the budget, but it is critical.

Once the test methodology is established, the scope of work is developed from the objectives of the
test program defined above. It is important that each objective is clearly addressed within the scope,
and that both the test work and the analytical work are defined properly. Figure 1 shows an illustration
that defines the sample size, test type, data produced, and objective of each component in a typical
metallurgical program.
5

Sample
Size Test Type Data Produced Objective

Tailings Process Model


Flotation TIF Design, Tailings Handling
Charact. Grinding
Pilot Plant Flotation
Moly Cleaning & Regrind
>10000 kg Moly Section Recovery, Grade
Separation Moly Separation
Tonnage Rate/ P80 Dewatering
Water Balance
Hardness Dewatering,
SPI, DWI, A*b, Wi, HPGR Reagent Selection
Tailings Transport Rmax,k = f(P80, pH,%sol) Operating Cost
Model
Rougher Tails Charact. Settling/Filtering/rheology/Tails Reagent/Lime
FT,FKT,SKT Consumption
Rmax/k = f(P80, pH) Energy Consumption
Ball/Liner
<1000 kg Cleaner Consumption
Conc Sample, ICP, NSR,
FKT,SKT
Column Ca, Jsl, Grade/Rec
Revenue Model
Locked Settling, U.A = f(P80, %Sol) Net Smelter Return
Cycle Tests
Dewatering Rmax,k = f(m,P80, Head Grade)

Rougher
Variability Model Validation Bankable Model

Block Model & Mine Plan Plant Optimization


Mineralogy
<20 kg Mineralogy =>
TPH, P80
Point Rmax, k
Hardness

Figure 1 Typical test program scope

What to watch out for

Possible disconnects between the team managing the test program and the team responsible for
handling and interpreting the data. It is critical that the modeling engineer be actively involved in
the test program supervision (remember, the key objective is a feasible process, not a table of test
results). This way there can be no disconnect between the test program and the engineering effort.
Lack of awareness of the strengths and weaknesses of the modeling and test program on the part
of the owners team. The key decision makers need to be made aware that not all process models,
and not all test programs, are created equal (they each have strengths and weaknesses). Today,
test programs are often done with more than one modeling methodology in parallel; this is
particularly true of grinding. This increases the cost but reduces the risks.
Failure to include the cost of data analysis, integration, interpretation, process modeling and
reporting. For modern test programs that involve statistical mass balancing of the test data, mineral
conversion, kinetics extraction, and modeling, this can amount to 10% - 25% of the cost of the test
program and 30% - 40% of the time required to compete it.
Failure to include a large contingency. Flotation test programs inevitably reveal unexpected
findings (thats why we do them) that require further work but simply cannot be foreseen at the
planning and budgeting stage. It is unwise to not allow for this in the budget. We recommend at
least 30% contingency in the budget, and up to 50% for a large program. These may seem like
high numbers, but in our experience it is easier to lose marks for going over budget than for not
spending contingency. On this point, the process metallurgist will likely find an ally in the Project
6

Director, who, if he or she is wise, has already included 10%-20% for undefined studies and will
have a 10% contingency back loaded in the project.

4 Sample Selection
The sample selection methods are a key aspect of the test program, and it is critical that this gets done
correctly. The authors routinely engage in critical reviews of prior test programs and in our opinion a
large fraction of current test programs are flawed because of sample selection problems.

Our preferred approach to sample selection is as follows. First, the geology or mine planning team is
asked to provide a list of all of the drill core intervals for each ore type. The information should include
drill hole ID, depth, geologic information, grades, and all other relevant information. This master list of
available core intervals should be filtered or reduced to a list of qualifying intervals by eliminating those
that are:

outside the pit limits, or outside the limits of the pit contours being considered for the test
program (sometimes this can be the first 10 years, or the first 15 years).
in waste, or are not in ore zones scheduled for processing (such as material below cutoff)
not available (the core recovery is too low, or it is degraded, or it has already been sampled for
previous test work)
clustered, causing an over-representation of one or more geographic areas of the deposit (this
is often caused by multiple drilling campaigns managed by different owners, or by fan drilling
from an exploration adit)

The remaining list of qualifying core intervals is then sampled to build the list of samples that will form
the ore domain composite or list of variability samples using a random number generator such that the
mean and standard deviation of the sample set matches that of the qualifying subset. Figure 2 shows
the cumulative frequency distribution of a qualifying list, the sample subset, and the list of samples that
were received by the lab for a recent test program. This kind of comparison is usually sufficient to pass
the gut check, but for more data-inclined engineers (or managers), a two sample t-test can be
performed to quantify the representativeness. If the data set is multi-dimensional and cross-correlated
(such when considering multiple assays), other tests are may be more appropriatethe chi-squared
test or the modified Kolmogorov-Smirnov test, for example (Corder 2009).

UGM1 - Cumulative Distribution of Cu Grades


100%
90%
80%
70%
% Less Than

60%
50%
Population
40%
30% Sample
20%
10% Received
0%
0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40
% Cu
Figure 2 Cumulative Distribution of Copper Grades in the Qualified Core List (blue), samples selected
(red), and samples received (green)
7

What to watch out for

Use of assay lab rejects. In the authors experience, the use of assay lab rejects should be
discouraged. In more than one case it was found that pyrite activation was a problem due to
oxidation of the crushed sampled obtained from rejects. As a worst case scenario, the authors
are aware of a least one real example of hydrolytic enrichment by excess sulfuric acid
generated by pyrite oxidation, resulting in the formation of mica or clay minerals (which are
known to impair flotation). At best, the use of reject samples opens the doors to criticism,
whether it is justified or not.
Improper sample support. For most variability samples it is agreed that bench height
composites offer the best compromise between error variance and granularity (assuming the
block height is equivalent to the bench height). For composite samples simple core intervals of
from 1 to 2 meters will suffice. Note that block caves require a very different methodology,
driven by the geologic or block model.
Improper clustering. The drill core collar intervals should be constant, or the core list should be
artificially declustered, and most geostatistics software packages can be used for this purpose.
This will prevent the sample set from being geographically biased.
Improper logging of the selected core intervals. It is very important that the sample list,
including drill-hole ID and down-hole depth, be accurate and included in all memos and reports.
Test work reports that do not include the sample source and selection criteria are not bankable,
by any definition. It is for this reason that many consultants and process engineers prefer to
participate in the sampling exercise personally, rather than depend solely on geology.
Inconsistencies between a properly selected random sample and the projected mill feed grade
for the period represented by the list of qualifying core intervals. If the mean head grades are
significantly different, this needs to be noted, with a complete explanation for the error, in the
report. There are often plausible explanations for thismost often it is a result of additional
drilling or reserve estimation completed after the initial sampling was performed.

5 Test Program
5.1 Assay methods
Proper selection of the appropriate assay methods is one of the most important considerations when
defining a test program. Assays can comprise 20% of the total test program costs for an ore with
relatively simple and predictable mineralogy, and up to 50% or more for ores with complex mineralogy,
precious metals, and/or multiple concentrate contaminants. Assay turnaround time can also account
for a large part of the duration of a test program.

When selecting the assay methods that will be used, consideration must be given to several points. A
complete review of the advantages and disadvantages of the various analytical chemistry techniques
will not be provided here; instead we will highlight some of the key points when choosing assay
methods. They are:

1. The assay methods used must reconcile with the assay methods that have been or will be used
to generate the block model and/or mill metallurgical control. It is preferable to use the same
procedures and standards, at least for the head samples, and it is important to establish by
round robin campaign that the various labs involved with the metallurgical program, mill control
(if any), and geochemical assays are unbiased and error-free. If the metallurgical assay lab is
8

biased with respect to the geochemistry assay lab, its not the end of the world. It just means
that the appropriate corrections need to be considered in the scale-up model.
2. Concentrate and intermediate mill samples will have higher grades and therefore will require
different procedures than those used for the feed and resource estimation. As a general rule,
samples greater than 10% copper, iron, or molybdenum will be performed using volumetric or
gravimetric methods. These will be higher costs. The acid strength or sample mass may also
be changed. High grade head samples also require a flag or descriptor of alternate assay
procedures were used.
3. Tails assays may require that a larger sample be digested in order for the detection limits to be
sufficient for the purpose. This is particular true of silver, which may require up to 5 grams to be
digested (for 0.1 ppm detection limits), or of gold, which may require up to 40 grams (for 0.01
ppm detection limits). These points should be discussed with the chief chemist. It is useful to
remember that that the assay lab works for the metallurgical engineer, and not the other way
around. Duplicate tails assays are a cheap but effective way to improve test reproducibility.
4. It is important to generate sulfur and iron assays, as these provide a method of estimating
residual pyrite (by sulfur balance) and iron oxides (by iron balance). If gypsum or anhydrite is
present, sulfate assays are required to develop a meaningful sulfur balance. Note that alunite is
not assayable by common HCl digestion.
5. Partial copper assays, such as copper soluble in dilute sulfuric acid, citric acid, sodium cyanide
solution, ferrous sulfate solution, etc., can be very effective for measuring secondary copper
sulfide species. They should be calibrated to quantitative mineralogy, as not all copper species
have the same equilibrium solubility in a given solution. They can be done in parallel or
sequentially. Sequential assays takes longer but are more precise.
6. Ensure that early on in greenfields programs, a well-documented set of assays has been
generated of the main minerals and all the possible deleterious elements that could give you
problems either in tailings as in the concentrate.

What to what out for

The metallurgist or test program manager who has not verified the assay lab procedures,
ensured that they are complete and consistent with those used for the geochemical program,
and incorporated a round robin test series to validate the procedures, has failed to perform a
critical part of his job.
Bid documents that do not specify the assays and assay procedures that must be used,
particularly for larger programs (FS level). This can be early warning that the project team is in
unfamiliar territory.
Test program quotes that dont specify the assays and procedures, the criteria for determining
which samples they get used on, and a description of the quality control and assurance
program. This can be a warning sign that the metallurgical lab are in unfamiliar territory.
Test program timelines that dont consider turnaround time for assays.
Assay program scopes that are incomplete. For sulfide projects, as a general rule, if the assay
scope doesnt include iron and sulfur, then it is likely incomplete, as there would be no way to
estimate the pyrite assay from stoichiometry even if the modal distribution of the primary
economic sulfides were known.
For large quotes, the lack of direct involvement on the part of the assay lab manager or chief
chemist can be an indication that the costs and turnaround times of the test program have not
been fully optimized. At the very least, the metallurgical lead should have a sit-down meeting
with the chief chemist to review the scope, cost, and timeline.
9

5.2 Mineralogy
The simple fact that revenue is driven by metal production, but flotation response is driven by
mineralogy, underscores the need for a proper mineralogical model that can be used to convert from
assays into minerals and back again. Most of the time, quantitative mineralogy (QEMSCAN, MLA, or
XRD) is sufficient to develop the mineralogical model for each ore type, but often optical microscopy is
also used. It is most common to use SEM-based analysis (QEMSCAN or MLA) of the feed and then
calibrated by reconciling the assays back-calculated from stoichiometry with those measured by the
assay lab. The same stoichiometry and calibrated mineral model can then be applied to the
concentrate and tails if there is no quantitative mineralogy information available. The process is shown
in Figure 3 (CuT, CuCN, and CuSA refer total copper, copper soluble in cyanide solution, and copper
soluble in citric acid, respectively)

Mineralogy
Assays Ag
CuT, CuCN, CuSA Au
CuFeS2

Fe
S
Mineralogy CuS2
Mo Model Cu5FeS4
Ag FeS2
Au For each Ore Type MoS2
As Cu3AsS4
Iron Oxides
Non Flotable Gangue

Figure 3 Typical mineralogy model

With the continuing reduction in the costs of automated mineralogical labs, it is becoming more
common to see quantitative XRD analysis performed on batch test concentrate samples, and
mineralogy (SEM-based or XRD-base, or both) on the tails. In this case the mineralogical model is
used primarily to normalize the various mineralogical estimates to the measured assays using statistical
or linear programming techniques. Note that this level of detail can be expensive, with simple modal
analysis costing on the order of $200 at a commercial lab at the time of writing.

Lastly, we should stress that the above discussion focuses on mineralogy for the purpose of flotation
testing and modeling. There are many other benefits of mineralogical characterization that have been
passed over in the short space allotted here. These include the benefits of understanding the
association, liberation, and deportment of the various economic and non-economic minerals, the
benefits of understand gangue mineralogy, etc.

What to watch out for

Any test program that does not consider and propose to model on the basis of mineralogy
should raise a flag. If a test program budget does not consider the cost of mineralogical
analysis, it should be assumed that mineralogy will be ignored. Mineralogy should be a
significant portion of the total program cost, typically between 5% (for simple and/or predictable
deposits) to over 20% (for complex deposits with multiple species of primary economic
minerals).
The delivery time for mineralogy is often longer than the other components of the metallurgical
test program. In the authors experience, it is usually wise to submit all samples for mineralogy
as soon as possible, so that these results are available when the test work needs to be
reported. In particular, gold deportment should be conducted as early as possible to direct the
flotation program when dealing with copper/gold deposits.
10

Be wary of mineralogy that is not reconciled with the chemical assays. Most mineralogy labs
perform this reconciliation as a matter of course, but the assays they used may be different from
those considered in the test program. Improper mineral models can lead to errors in the
flotation test mass balance algorithms, kinetics extraction, and process modeling.
Reports that do not clearly describe the methods and assumptions used to convert between
assays and minerals. Once established, the mineralogy models needs to be consistent across
all stages in the test program. For example, assuming that secondary copper sulfides are
primarily bornite during the test program, but primarily chalcocite during modeling, will lead to
errors in the modeling of pyrite and iron oxides (because of the differences in stoichiometry).
Insist that the mineralogical lab report the source of the calibration information if using
QEMSCAN or similar data. It particular, it should be stated whether the library was built from
samples of the deposit itself or if the library was based on other similar deposits. For example,
we are aware of cases in which diurleite and not chalcocite was the predominant supergene
copper sulfide mineral, yielding significant differences in the estimated copper deportment. This
can be significant for large concentrators.

5.2.1 Flotation Tests


There are lots of ways to conduct a flotation test, and not all metallurgists are in agreement on what
should constitute the proper procedure. For this discussion we will therefore focus on those points with
which, we hope, most of us are in agreement.
5.2.1.1 Kinetics Tests
The kinetics test is the workhorse of flotation testing. It is the only means by which the necessary
residence time in an industrial bank can be calculated. Before a lab can begin churning out large
volumes of flotation kinetics, however, several details must be seen to.

For starters, the flotation cell needs to be calibrated. This procedure involves ensuring first that the air
flow rate is measured and the measuring device is calibrated. Measurements are taken of the cell
volume (actual and effective) and cell surface area (actual and effective). If possible the impeller speed
is checked with a tachometer. Then several tests are performed at different air flow rates and with
different impeller speeds. The flotation kinetics at each different condition are measured and the best
condition is selected; i.e., that which maximizes the flotation efficiency. Once the cell has been
calibrated, the air flow rate, impeller speed, and cell volume should be held constant for the remainder
of the test program (at least for the kinetics tests).

The next step is to identify the best reagent suite to use for the test program. This is usually done on a
composite sample representing the first five or ten years of ore. Factorial design, Taguchi or other
experimental design methods can be employed. It is important to check both reagent type and dosage,
and often pH is varied during this step as well, to investigate possible interaction between pH and
reagent type (sometimes grind will have an impact on reagent response as well)

Once the reagents are selected, establish the baseline conditions (usually from previous studies) and
then the primary rougher kinetics tests are performed. As a minimum, these should include, for each
ore type:

Flotation at different grind sizes


Flotation at different pH
Flotation at different percent solids
Flotation at different pulp oxidation-reduction potential (if necessary)
11

The results of the rougher tests are used to select the conditions (see next section) that will be used for
the rougher variability testing and the fabricationthrough a bulk flotation stepof sufficient rougher
concentrate for subsequent cleaner kinetics testing.

The rougher variability tests are rougher kinetics tests performed on randomly selected block or bench
composites representing the distribution of hardness, head grade and mineralogy in each ore type.
Typically there are at least 30 individual variability samples tested, in order to generate sufficient data
points to infer the characteristics of the underlying distribution. These tests are used to:

Evaluate relationships between mineralogical parameters and develop a mineralogy model that
can be applied to the assays in the block model
Evaluate changes in kinetics due to changes in the slope of the rougher feed size distribution
Evaluate changes in kinetics due to differences in head grades
When done at a constant grind time, they provide knowledge on variation of ore hardness.

The cleaner kinetics test series is similar to the roughers; and includes flotation at different pH, regrind
size, as well as other conditions.

The cleaner tests add considerable cost to the overall test program, because they generally require a
much larger sample of rougher feed just to generate sufficient quantities of rougher concentrate for
kinetics testing. However, they are just as critical to the overall success of the test program as the
rougher tests. It is easy to design a roughing stage, because they are usually configured in open
circuit. But the circulating loads in the cleaners and the sensitivity of the regrind circuit performance to
changes in head grade, mineralogy and circulating loads can complicate things in the cleaners. For
this reason, cleaner kinetics tests are an essential part of every test program, and they should be done
on every ore type.

5.2.1.2 Open and Closed Cycle Tests


Locked cycle tests are multi-stage, multi-cycle tests designed to provide a somewhat more faithful
reproduction of the industrial flow sheet (Agar 2000). They provide an estimate of the final grade and
recovery for fast floating mineralsif the locked cycle test achieves good recoveries and grades, then it
will also be possible to design a process plant capable of achieving good recoveries and grades.
Locked cycle tests cant confirm, however, if a given circuit design will be capable of delivering good
recoveries and grades. This is because locked cycle tests suffer some serious drawbacks. These
include:

They are difficult to control, particularly for ores with low head grades or slow-floating minerals.
They can require a large number of cycles to reach equilibrium for slow floating minerals, such
as molybdenite.
They do not provide kinetics information, so they cant be used to estimate the required size of
cleaning stages.

Because of the low percent solids in the final stages of locked cycle tests, and the higher water
recovery achieved in those concentrates, caution should be used when extrapolating the circulating
loads measured in the final cycle of a locked cycle test to those expected in a plant.
12

Open cycle tests are even more problematic, because the lack of recirculating tails in the cleaner circuit
makes it difficult to conclude anything with respect to the cleaner circuit recovery even for a well-
executed test. For this reason, they are often used only for qualitative or scoping work.

5.2.1.3 Column Tests


The column flotation tests are performed using a small diameter laboratory batch apparatus. Because
the column cell is unique in that it effectively eliminates entrainment, the stage grade-recovery curve is
independent of the feed grade. Hence, the grade-recovery curve of the laboratory column, even when
fed with a reground rougher concentrate (the grade of which is usually much lower than that expected
for a typical industrial column cleaning application), is directly scalable to the plant. The test produces
a sample that represents the industrial concentrate, and the grade-recovery curve can be used to
validate the circulating loads derived from the plant model. The samples of concentrate generated can
then be used for settling, rheology and filtration tests.

What to watch out for

All lab test programs should capture and recycle the water used for flotation tests (once
standard reagent conditions are identified) so that the water reclamation ratio of the lab matches
that of the plant. This will reduce costs of water disposal, improve the quality of the results
because the lime and reagent consumption will be closer to plant conditions, and provide a
more representative sample of process water (for environmental testing).
The reagent dosage in a lab can be quite different from that in a plant environment, particularly
for frother and oily collectors (such as diesel). This is because there is a large amount of
recirculation of reagents in plant systems that may not be present in a lab, particularly during the
early tests before the water recycle system has reached steady state
The bulk of the flotation work should be with standard kinetics tests for both rougher and
cleaner. Kinetics tests should have an independent head assay (to check for sampling errors),
and an independent feed screen analysis (to check for sample preparation errors), and if
possible, duplicate tails assays (to improve reproducibility).
Always ask for the concentrate wet and dry weight to be reported, so that water recovery can be
estimated.
Beware of so-called eye-ball standard kinetic tests, or tests for which the air rate and froth
removal rate are varied according to the whim of the technician. This kind of pseudo-kinetics
test is not normally useful for kinetics modeling (it is useful, however, for generating a sample of
rougher concentrate for further work). Another example is the use of very small quantities of
collector in fresh water, yielding collector concentrations far lower than what can be properly
controlled in a plant. While this approach can give high concentrate grades for a given
recovery, it is not usually useful for modeling and design.
Watch for incomplete test report sheets. A good rule of thumb is to ensure that all tests are
reported with the cell volume, impeller diameter and speed, air rate, pH, pulp potential, pulp
temperature, percent solids, reagents, conditioning time, grinding time, and concentrate and
tails percent solids. The grinding media, mill size, mill speed, charge, charge type, and charge
materials of construction (steel, stainless, or ceramic) should also be noted.
When testing flotation variables, ensure you test a wide range.
We strongly encourage the testing of four points per variable in order to define the response
over the full range of the parameter. Remember that three points define a curve, and the fourth
one confirms it.
13

Be cautious in doing cleaner tests without any reagents. This may be possible in the plant, but
remember: by the time the cleaner feed stock is reground, the reagents may have degraded. It
is prudent to add about 10% of the rougher reagents to cleaners.
Be cautious with test programs that rely too much on locked cycle tests, or open cycle tests.
Locked cycle tests cannot be used to design a circuit; hence it is usually unnecessary to
perform more than one or two locked cycle tests per ore type. This should be more than
enough to generate the concentrate quality information and all the tailings sample required for
tailings and environmental characterization.

6 Process Modeling
Recent years have seen much growth in the modeling and interpretive component of current flotation
test programs. While it is true that some owners still opt to split up the test program component and
modeling component, advanced modeling and analytical methods are being increasingly integrated with
the test program in order to reduce costs, time requirements and risk. The approach described in this
section is the method favored by the authors, but there are some variations in use by others.

Some degree of engineering and interpretation is involved in almost every step of the test program;
Figure 4 shows an example of the work involved in developing to completion a feasibility level flotation
test program. In the words of one of the authors, the figure looks like something the US Department of
Agriculture might have produced; for this we do apologize. In the following sections we describe ad
give examples of the key aspects of the process. Note that the work process can vary from project to
project, and portions of it can be iterative as you progress through the various stages of the project
development effort.

6.1 Modeling & Interpretive Steps

6.1.1 Kinetics Test Interpretation


All kinetics tests require interpretation and kinetics extraction. The method of interpreting depends
upon the kind of model used, but the main steps are usually the same. They are:

1. Ensure that the assays and mass are balanced using an accepted mass balancing method.
2. Convert the balanced assays into balanced minerals, calculating the assay of gangue by
difference. The gangue assay can be checked by XRD if the budget allows.
3. Calculate the kinetics parameters for each mineral, including gangue, using a least squares
fitting method in combination with some kind of batch flotation model. The particular model
used by the authors is described elsewhere (Amelunxen & Amelunxen, 2009).
14

Rougher kinetics series Rougher Variability


(P80, pH, %Sol) Tests
Qa/Qc Repeats (kinetics) Error Propagation
Qa/Qc Repeats (assays) Study
Add head grade &
mineralogy to
Rougher Kinetics Multi
equations
Regression
Rougher Optimum Monte Carlo
Conditions Simulation Block Model
SIMULATIONS
Equations
Bulk Float to produce Tailings
Rougher Conc Characterization
MINE PLAN
Tonnes per ore type, head
FLOTATION PROCESS grades, mineralogy
MODEL
Cleaner kinetics series
(P80, pH, %Sol)
Prelim Water
Qa/Qc Repeats (kinetics) Balance COST MODEL
Qa/Qc Repeats (assays) (CAPEX, OPEX, TC/RC,
freight, penalties, etc.)

Cleaner Kinetics Multi REVENUE MODEL


Regression (metal prices, price
Cleaner Optimum Locked Cycle Test participation, royalties, etc.)
Conditions Model
Validated Model
Optimized Process Configuration
Column Test Locked Cycle Test Plant Model Validation Production Plan
Cost & Revenue Forecast
Mass & Water Balances for each variability sample
Process Flow Distribution
Process Design Criteria
Error Range & Confidence Level of rec, grind
Sample of representative, saleable concentrate

Figure 4 Typical model and test work components. Blue boxes are modeling steps and brown boxes are
testing steps.

6.1.2 Rougher Multi-Regression


The rougher kinetics multi regression step consists of developing an equation that predicts the
collection zone kinetics parameters of each mineral as a polynomial function of the P80, pH, pulp
percent solids, or other test variables considered. A separate equation is used for each mineral; it is
often (but not always) of the form:

Figure 5 shows the predicted versus actual scatterplots for the Rmax and K parameters for an ore type.
The kinetics model is used to rapidly calculate the flotation kinetics for all the minerals for any given set
of operating conditions. This is a necessity for the subsequent process optimization exercise.
15

Rmax Ag K Ag
Rmax [%] Regression Scatterplot Rmax As
K [1/min] Regression Scatterplot K As
100 Rmax Al K Al
4.0 K Ca
Rmax Ca
90 Rmax K KK
3.5 K Mg
80 Rmax Mg
Rmax Mn K Mn
3.0 KP
70 Rmax P
Rmax Ti K Ti
60 2.5 KV
Rmax V
Actual

Actual
Rmax CuFeS2 K CuFeS2
50 2.0 K Cu5FeS4
Rmax Cu5FeS4
40 Rmax CuS K CuS
Rmax CuOX 1.5 K CuOX
30 Rmax MoS2 K MoS2
Rmax CaMoO4 1.0 K CaMoO4
20 Rmax ZnS K ZnS
10 Rmax PbS 0.5 K PbS
Rmax CaSO4 K CaSO4
0 Rmax FeS2 0.0 K FeS2
0 10 20 30 40 50 60 70 80 90 100 Rmax Fe3O4 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 K Fe3O4
Rmax Gangue K Gangue
Model [Rmax = f(pH, %Sol, P80) Ideal Model [k = f(pH, %Sol, P80) Ideal

Figure 5 rougher kinetics multi-regression scatterplot for Rmax and K

6.1.3 Rougher Optimization


The rougher kinetics multi-regressions described above are used to determine the optimum rougher
operating conditions using a response surface methodology (RSM) procedure. The method consists of
performing a Monte Carlo style simulation of pH and pulp percent solids (or any other conditions) for
the baseline P80. Note that the optimum grind is not defined yet because defining the grind is an
economic decision that requires capital and operating cost modelsthis will be discussed below.
When multiple economic minerals are present (such as copper, molybdenum and silver in the example
shown above), then the optimum conditions must be evaluated on the basis of the gross specific
revenue, or dollars per tonne of ore assuming as a first pass typical cleaner recoveries for each
mineral. Figure 6 shows an example of the specific revenue surface plotted against pH and P80 for a
single ore type.

The best conditions are then used to produce the bulk sample of rougher concentrate for subsequent
cleaner tests. In this way the cleaner concentrate is more representative of that which would be
produced in the plant.

6.1.4 Cleaner Multi Regression


The cleaner multi regression procedure is identical to that performed for the rougher kinetics tests. The
purpose is to express the cleaner collection zone kinetics as polynomial functions of the cleaner pH,
regrind P80, and pulp percent solids.

6.1.5 Adding Head Grade & Mineralogy to Rougher Models


Once the regression models have been derived for each ore type composite (see Section 6.1.2), the
results of the variability test work is incorporated by introducing an eighth constant, c 8, as follows:
16

Figure 6 Revenue ($/tonne) Surface plot of pH and P80

For those ore bodies that contain an imposing oxide or acid soluble copper it may be necessary include
a term for acid soluble species. Often a fixed grind time is used for the variability samples, resulting in
variation in the P80 and slope of the size distribution curve. In this case, a term to correct for the slope
is also added to the regression equation. The regression constants c3 through c7 are then held constant
and the values for c1, c2, c3, and c7 are fitted1 to the variability and composite test results such that sum
of the squared residuals of the Rmax values, k values, and equivalent recoveries are minimized. The
result is a unique equation for each mineral that expresses the rougher collection zone Rmax and K
values as a nonlinear function of the grind size, pulp pH, slurry percent solids, head grade, and slope of
the size distribution.

6.1.6 Cleaner Optimization


The purpose of the cleaner optimization is to identify the conditions for each ore type that yield the best
metallurgical performance. Unlike the roughers, the regrind size is included as a variable in the cleaner
circuit optimization because the capital and operating cost of the regrind circuit is generally negligible
compared to its impact on the revenue (unlike the primary grind size is for the roughers).

1
If the variability tests are conducted for a fixed grind , then c2 and c3 are also held constant.
17

Because the cleaner optimization is the last piece of information required to start simulating the process
plant, the optimization is done a bit differently. For the roughers, the best operating conditions are
defined by maximizing the specific revenue, which in turn is calculated on the basis of rougher
recovery. In selecting a stage configuration, the modeler needs to take into account the hydrodynamics
of the cell, and ensure that it conforms to industry best practices. This requires knowledge of key
performance indicators such as slurry velocities within a cell, overflow lip loading constraints, and
carrying capacities. The number of cells in a row of cells needs to be economically optimized while
maintaining a reasonable degree of froth pull, (referred to by modelers as the froth recovery). For the
cleaners, the plant is configured with a base case model (the configuration of which will be optimized
later) and simulations are conducted at different combinations of operating pH, percent solids, and
regrind size. For each node in the simulation matrix, an iterative optimization routine is used to search
for the point on the curve of grade versus recovery that yields the highest simple net revenue (defined
as gross sales revenue minus freight costs, treatment costs, and any refining and penalty charges).
The node with the highest optimized net revenue is the node with the best operating conditions for the
cleaners. These conditions are then used for modeling and carrying out the locked cycle test, as
described below.

6.1.7 Locked Cycle Test Model


The locked cycle test model is a kinetics model similar to an industrial plant model. It is based on the
same principals of first order kinetics and separate sub models for collection, froth recovery and
entrainment. It is used to derive a mass balance of the locked cycle test prior to commencing the test,
so that the concentrate weights (wet and dry weights) can be controlled during the test. This provides
the technician a much greater degree of accuracy in achieving the target recovery and concentrate
grades when performing the test.

6.1.8 Model Validation


Model validation is performed by comparing the process model to a known process. If the model
predicts faithfully the recoveries and concentrate grades of not just one, but all minerals, then it can be
considered valid. If the model is configured with the measured collection zone rate constants and
maximum recoveries and no artificial correction factors are introduced, then, because it is a
phenomenological model, it can be legitimately extrapolated to different process configurations and
equipment sizes (Amelunxen and Amelunxen 2009). At the greenfields stage, locked cycle tests and
pilot plant tests are used to validate the model. For brownfields, plant audits can also be used.

6.1.9 Error Propagation Study


Perhaps the most common question we receive when submitting a process performance forecast is,
How accurate is it? There are two aspects the question. The first is the question of model bias and
scale-up error, or accuracy. This question can only be answered by performing model validation
studies around existing flotation circuits and comparing the modeled and actual process performance
(the authors routinely engage in these types of studies, but for space reasons we cannot summarize
them here).

The second aspect relates to the precision of the resulting recovery, grade, or mass balance estimates
predicted by the model. This is the purpose of the error propagation study: to draw error bars around
the estimated recoveries and grades derived from the test program. Because the modeling and scale-
up procedures are complex and nonlinear, it is preferable to use Monte Carlo methods applied to
measured error variances, rather than an analytical solution.
18

Scale-up error must be assessed in the context of the entire test program rather than the reproducibility
of a single flotation test. In other words, we are interested in the standard error of the mean recovery,
not the standard deviation of a test parameter. The standard error of the mean is defined by:

(Eq. 1)

Where: S is the standard error of the mean,


is the standard deviation, and
n is the number of independent measurements or degrees of freedom of the parameter.

The purpose of the Monte Carlo simulation study is to translate the standard error of the mean kinetics
parameters into the standard error of the projected flotation circuit recovery. This can be done either
for a single grade recovery curve or for the entire mine plan. The general procedure is as follows:

1. From repeat assays performed on different samples during the course of the test program,
develop an error model that expresses the assay error variance as a function of the assay
value.
2. Select a kinetics test. Assuming that the measured assays are the true assays, use Monte
Carlo methods to simulate error for each assay. At this point error associated with flotation test
repeatability is included by considering the variance of concentrate mass recoveries derived for
repeat tests on the same sample. Add the simulated error the measured assay to produce a
simulated kinetics test.
3. Simulate many kinetics tests and calculate the resulting kinetics for each mineral. The variance
of each kinetics parameter is the error variance due to uncertainty in the assays and flotation
test.
4. Convert the error variance of each kinetics parameter into variance of the mean kinetics in order
to reflect the degrees of freedom remaining after the kinetics regressions are performed. This is
done using Eq. 1.
5. Perform a second Monte Carlo simulation study using the process simulator. Simulate the error
of the mean kinetics parameters for each mineral and model the process using the simulated
kinetics. In order to compare apples with apples the process must be optimized for each new
simulated set of kinetics parameters.

The result gives an estimate of the dispersion about a grade recovery curve or recovery estimate, as
shown in the following figures.
19

Figure 7 Simulated error in estimated grade-recovery curve due to sampling & assay uncertainty

Figure 8 Simulated error in copper recovery plan due to sampling and assay uncertainty

6.1.10 Block Model Equations


The block model equations consist of a series of polynomial equations that express the plant recovery
and concentrate grade of a given element or mineral as a function of the head grade, ore hardness,
mineralogy, or any other variable that is available at the block model level. They are input into the
block model and used for mine planning and reserve estimation purposes.
20

6.1.11 Mine Plan


The Mine Plan is prepared by the client or owners team. It usually consists of a series of time
periods usually years, but often the first several years after project startup shorter time periods, such
as quarters or months, are used. For each time period the blend of ore types is reported, including the
tons of each ore type scheduled for processing, the head grade and any available mineralogical
information, and the average ore hardness of each.

6.1.12 Cost Model


The cost model consists of equipment capital costs including installation and indirect costs, operating
costs including reagent and energy consumption, and concentrate treatment and refining charges,
shipping and freight, and penalty charges.

6.1.13 Revenue Model


The revenue model is based on either a typical smelter contract or a client-specified metal price
schedule. It usually includes any royalties or price participation agreements and downstream losses.

6.2 Deriving the Optimum Circuit Configuration


The optimum circuit configuration is defined as the circuit configuration that yields the highest net
present value for a given mine plan and smelter contract (or equivalent).

The first step is to determine the correct grind size. This is done by selecting a grinding circuit and then
simulating the process plant performance over the entire mine plan. For each period in the mine plan,
the conditions are optimized by searching for the operating conditions that yield the highest possible
NPV for the corresponding capital cost, operating cost, and grind size. Then the installed ball mill
power is increased, the capital and operating cost recalculated, and the new NPV is calculated. The
process is repeated over a series of cases of increasing grinding circuit size, cost, and installed power
resulting in a finer grind. The point at which the increased cost is no longer justified by the increased
revenue is the optimum grinding circuit size, and the resulting grind size is the optimized nominal grind
size. Note that for each grinding circuit scenario the tonnage is also optimized, and operational
strategies such as increasing tonnage at the expense of grind can be considered in the simulation
without any difficulty, provided the mine has the capacity to deliver. Figure 9 shows the results of such
a study for both a SAG mill circuit and an AG mill circuit. The optimum grind is determined by
optimizing cash flow over the entire mine plan (10-years, in this case). For this example the grind is
approximately 140 for the SAG mill case.
21

NPV10 vs P80
$3, 000,00 0

NPV10
$2, 900,00 0

$2, 800,00 0

$2, 700,00 0

$2, 600,00 0

$2, 500,00 0

$2, 400,00 0

Fully Autogenous Grinding


$2, 300,00 0

Semi-Autogenous Grinding
$2, 200,00 0

$2, 100,00 0

$2, 000,00 0
P80 (microns)
60 80 100 120 140 160 180 200 220

Figure 9 Determination of optimum grind over a 10-year mine life, two grinding circuits

Once the economic optimum P80 is determined, the roughers are sized, and the strategy is similar.
Additional roughing capacity is added sequentially until the net present value of the optimized mine plan
is maximized, after which the additional capital and operating cost associated with increased rougher
retention time is no longer offset by the increase in recovery.

NPV vs No. of Cells


$2, 350,00 0
NPV10

$2, 300,00 0

$2, 250,00 0

$2, 200,00 0

$2, 150,00 0

$2, 100,00 0

$2, 050,00 0

$2, 000,00 0

Rougher Optimization
$1, 950,00 0

Scavenger
$1, 900,00 0

0 1 2 3 4 5 6 7 8 9
No. of Cells (160 m3)
Figure 10 NPV-optimized rougher and scavenger configuration

With the roughers optimized, the cleaners are next. The first cleaner, cleaner-scavenger and other
cleaning stages are optimized in the same way. Different circuit configurations can also be evaluated,
all on the basis of optimized NPV.

What to watch out for


Failure to use flotation kinetics rather than simple recoveries and concentrate grades to forecast
plant performance. There are far too many differences between the flotation conditions in the
lab and the flotation conditions in the plant to consider empirical scale-up viable. This is
especially true of the lower grade deposits being developed in modern times.
22

Be wary of programs that conclude that a single optimum pH, regrind size, or % solids is
adequate. The optimum flotation conditions are related to the P80, head grades, and blend of
ore types that are feeding the concentrator at a given moment or in a given production period. If
this is not recognized, then it is likely that there is still considerable value (in the form of NPV)
that can be unlocked. There are tools available to do this.
Take the locked cycle tests with a grain of salt. Good results in a locked cycle test usually
indicate that the plant can do even better. Bad results usually indicate nothing. If good results
are required (for example, to demonstrate feasibility, validate a model, or meet a stage gate
requirement), then we recommend modeling the locked cycle test and performing a mass-
controlled locked cycle test to the specifications dictated by the model. Dont be afraid to
engineer the locked cycle test to give the desired result.
Insist that the model be validated, either through open cycle, locked cycle, pilot, or industrial test
programs. If a model is not validated, then the study is probably not bankable (in the authors
opinion). It is not difficult to validate a model. (Amelunxen and Amelunxen 2009)
Beware of test programs that fail to include operating costs, capital costs, and revenues in some
form or another. A process design is not optimized if the costs and revenues have not been
fully accounted for, and there is no way of getting around this. If competing process design
alternatives are not compared on the basis of optimized operating points, then we are
comparing apples and oranges.
Lastly, once the test program has been completed, probably at great time and expense, it is
critical to ensure that the data integrity is protected and preserved for future use. There are
plenty of proper storage databases and formats. Note that test procedures and raw data should
be preserved, as well as modeled and simulated interpretation.

7 Conclusions

This paper has presented an overview of the various components of modern flotation test programs.
Hopefully it has become apparent that developing a complete and successful test program requires a
solid understanding of modeling, analysis, engineering, and cost estimation. For example, the process
optimization step described above cannot be done unless the right operating conditions were tested
and a regression model was developed to relate the kinetics parameters to those operating conditions.
Furthermore, the test procedures need to be compatible with the kind of process model being used to
perform the optimization.

Nevertheless, it is often the case that a test program is managed by a project metallurgist or consultant
who reviews the results, produces a report and then hands the results over to the engineering
company. It is then the job of the engineering group to develop the cost models and derive the
economic optimum flow sheet. It should be obvious by this point that we think this is a mistake, as
there are significant advantages that can be gained by integrating the engineering directly into the test
program (or at the very least, integrating the engineering companys metallurgist). These include:

By using a modeling approach, rather than a trial-and-error approach, the work flow can be
streamlined because tests can be conducted in parallel, rather than having to wait for assays
every time test series is performed to investigate a process variable. Significant time savings
can result.
Considering costs and revenue when defining the best rougher and cleaner conditions can help
avoid having to repeat some test work if better economic optimum conditions are found later on.
23

Considering optimum conditions when performing the locked cycle tests, column tests, and
tailings characterization tests also helps to avoid mistakes. Often the tailings and concentrates
of these tests are used for subsequent tests, and any change in the test condition might
undermine the results. One comment we frequently hear regarding tailings rheology or tailings
acid generation tests is that for some reason, the tailings size distribution tested does not
match the design criteria for the industrial plan. This often happens when the optimum grind is
identified after the test program has already been complete.
By producing an optimized design before the engineering effort gets underway, costly trade-off
studies can be avoided (obviously this is only a benefit if youre not an engineering company
selling costly trade-off studies)

8 Works Cited
Agar, G. "Calculation of locked cycle flotatin test results." Minerals Engineering 13 (2000): 1533-1542.
Amelunxen, P, and R. Amelunxen. "Methodology for using laboratory kinetic flotatni parameters for
plant design and optimization, Part 2." Procemin 2009. Santiago, Chile, 2009.
Amelunxen, R, and P. Amelunxen. "Methodology for executing, interpreting, and applying kinetic
flotatino tests in scale-up, Part 1." Procemin 2009. Santiago, Chile, 2009.
Baumgartner, R, et al. "Building a geometallurgical model for early-stage project development - A case
study from Canahuire epithermal Au-Cu-Ag deposite, sourthern Peru." The First AUSIMM
International Geometallurgy Conference. Brisbane: AUSIMM, 2011. 53-59.
Corder, G., and Dale, I. Nonpartametric Statistics for Non-Statisticians. Wiley, 2009.
Montoya, P.A., et al. "Geometallurgical modelling techniques applicable to prefeasibility projects - La
Colosa case study." The First AUSIMM International Geometallurgy Conference. Brisbane:
AUSIMM, 2011. 103-111.

S-ar putea să vă placă și