The
Relative Bias Errors of Gravimetric Fire-Assaying Practice for Platinum-Group
Elements in Bushveld Merensky
NO Lotter 1,
CT O’Connor 2 and I Clark 3
April 2000
SUMMARY
The evaluation of PGE in Rustenburg Merensky ore and mineral
process products has been studied with potential
biases as a specific issue. It was found that the lead-collection fire-assaying method,
which is widely used in
Introduction
Numerous references exist for the fire-assaying of gold-bearing samples in the South African mining industry, some of which are noted here [Eissler, 1896; Adamson, 1972, Lenehan and de L. Murray-Smith, 1986]. Most of these describe a pre-concentration fusion step followed by cupellation and gravimetric determination of the gold prill after appropriate parting for silver. The adaptation of this practice to the gravimetric determination of the PGE in evaluation of the Merensky and UG2 ore types of the Bushveld Complex, South Africa, found favour because it was more rapid than the lengthy classical wet analytical technique [Lenehan and de L. Murray-Smith, 1986].This fire-assay technique for the PGE-bearing ore types and metallurgical process products differs from the equivalent method for determination of gold in Witwatersrand gold-bearing ore types for several reasons. These include the alteration of the flux to suit the more basic nature of the Merensky host rock, and resetting the fusion times and temperatures of the two cupelling stages.
Typical
Fire-Assaying Practice for Gravimetric Determinations on Samples of Merensky Reef
Fire assaying is commonly used in the determination of trace
concentrations of precious metals in ores and mineral process products. This is because it pre-concentrates the
precious metals from the sample by fusion with a flux. This produces a lead button, which collects
the precious metals present, leaving a slag, which is
separated from the lead button.
The flux used typically contains litharge, sodium carbonate and
borax. A reducing agent, usually maize
(corn) meal, is used to react with the litharge and
produce lead metal and carbon dioxide.
Occasionally potassium nitrate is added to the
flux to oxidise any sulphide minerals present in the sample. This is to counteract the naturally reducing
nature of the base metal sulphides.
Lead
Fusion Stage
The purpose of
this stage is to quantitatively pre-concentrate the precious metals into a lead
button. The weighed sample is mixed with the flux and the mixture placed in a
crucible. Batches of weighed sample and
flux are prepared and placed in a series of crucibles. This batch of crucibles
and their contents is heated in a pre-heated furnace,
which is operated under controlled conditions.
The rate of heating and final temperature during the
fusion are critical controls. The
initial furnace temperature is 11000 C. As the batch of crucibles is loaded into the
furnace, the temperature drops to approximately 8500 C, and rises
again to 11000 C over a 45-minute period. On completion of the fusion
the crucibles are unloaded singly from the furnace, and the crucible contents
poured into an inverse conical mould.
The fusion products settle and solidify as they cool. After cooling the
mould is inverted and the lead button removed from the slag. The efficient recovery of the PGE present in
the sample to the lead button, which is critical to the reliability of this
method, is dependent on appropriate comminution of the sample prior to fusion,
thorough mixing of the sample and flux, and temperature control during the
fusion. The mass of metallic lead is
also an important factor, which needs to ne
controlled during the adjustment of flour or maize (corn) meal and nitre (potassium nitrate).
Volatile losses have also been discussed, which
have been treated or limited by the use of lids on the fusion pots [Munro,
1993]. The typical flux compositions for
the fusion of samples of Merensky ore and mineral process products are
summarised in Table 1 [Williams, 1994].
Table 1 - Typical
Lead Fusion Flux Composition
|
Flux Constituent % By Mass |
|
Flotation
Concentrate Samples |
|
Sodium Carbonate |
35.1 |
31.5 |
|
Borax |
26.3 |
23.6 |
|
Silica |
8.8 |
15.7 |
|
Litharge |
26.3 |
23.6 |
|
Maize (Corn) Meal |
3.5 |
3.2 |
|
Potassium Nitrate |
0.0 |
2.4 |
Low-Temperature
Cupellation Stage
The purpose of
this stage is to quantitatively remove the lead from
the precious metals. The lead buttons are placed individually on preheated cupels, in a
cupellation furnace at 900-1100 degrees centigrade. The cupels are made
from calcined magnesite. These cupels
absorb the molten lead and leave behind a prill of precious metals. Some of the lead is
volatilised during this process. The product of this stage is called a ‘low-temperature prill’, or ‘LT prill’.
High-Temperature
Cupellation Stage
The purpose of
this stage is to quantitatively burn off the volatile precious metals Ru, Os,
and Ir, since partial losses of these occur in the LT cupellation stage. This leaves a ‘high-temperature prill’, or
‘HT prill’, after HT cupellation at 1300-1350 degrees. This also incurs secondary losses of the four
remaining precious metals, Pt, Pd, Au and Rh.
Gravimetric
Determination
The HT prills
produced from the above processes weigh between 1 mg for flotation concentrate
samples, and 0.1 mg for low-grade tailings samples. The handling of these prills is therefore a
delicate matter, as breakage of the prill during handling negates all of the
foregoing effort to quantitatively produce a prill for
weighing on a microbalance.
Assay
Replicates
Quadruplicate
determinations are the standard practice in production metal-accounting
laboratory operations. The arithmetic
mean of these replicates is calculated after
inspection of the replicate values for precison; this introduces a necessary
quality control process that, unless appropriately formulated with knowledge of
the complex mineralogy and skew distributions, will bias the estimation of mean
grade. This matter is outside the scope
of this paper, and will be separately reported. The experimental work reported here, however,
used multiple replicate determinations per test condition in order to overcome
this difficulty.
Correction
Factors
In routine
fire-assaying of production metal accounting samples which have a processing
and analytical history, empirical correction factors are developed which
correct for the cumulative minor loss of Pt, Pd, Au and Rh in the LT and HT
cupellation stages. The factor is only applied to the HT prill mass. LT prills are an
intermediate stage of the process and are not weighed. These factors are in the order of 3% in ore
samples, 7% for tailings samples, and 1% for concentrate samples. Nickel collection is used
to calibrate these factors opposite multiple replicate fire-assay
determinations of the sample samples.
The procedure of nickel collection is outside the scope of this paper,
except to note that this method produces a more complete determination of the
precious metals [Robert et. al., 1975, Munro, 1994]. In this paper the
units of PGE quoted shall be ‘4 element uncorrected’, i.e. the sum of Pt, Pd,
Au and Rh, without correction factors applied.
Description of the Main Mineralogical
Features of the Merensky Ore Type
The Bushveld
Complex is an extensive geological system in the Northwest and
Specifically
the Merensky Reef commences with a 1cm thin chromitite layer often resting on
anorthositic norite; the chromitite layer is well defined
and is overlain by a pegmatoidal feldspathic pyroxenite. This pegmatoidal layer is generally some 25
cm thick, and comprises subhedral- to euhedral- orthopyroxene crystals with
less abundant plagioclase feldspar. A
second 1 cm chromitite layer occurs at the top of this coarse-grained unit, and
demarcates the top of the geological entity called the Merensky Reef of
Rustenburg Platinum Mines.
The regional
features of this system show that variations in the Merensky Reef are emulated in the lower UG2 system in the same vicinity
[Kinloch, 1982]. This variation is
characterised by the relative abundance of the discrete platinum group minerals
(PGM). In particular, isoferroplatinum
(Pt3Fe) and the platinum-palladium sulphides exchange as the
dominant discrete PGM form at a given site, depending on the geological history
of a particular area of the Bushveld. In
areas where secondary activity such as a volcanic intrusion occurred,
isoferroplatinum is dominant over the platinum-palladium sulphides. In Rustenburg Section of Rustenburg Platinum
Mines, the platinum-palladium sulphides are dominant over isoferroplatinum.
In terms of
the general base metal sulphide mineralogy, the Merensky ore type hosts
chalcopyrite, pentlandite, and pyrrhotite with minor pyrite. The pentlandite and pyrrhotite are enriched by PGE in solid solution to varying
degrees. Also,
blebs of discrete PGM occur as intergrowths with pentlandite and
pyrrhotite. For Rustenburg, the relative
abundance of the PGM is described in Table 2, and the
enrichment of the base metal sulphides, in Table 3.
Table 2 - Relative
Abundance of the PGM at Rustenburg
(After Kinloch, 1982)
|
Platinum Group
Mineral |
Volume Distribution
as % of all PGM |
|
iso-Ferroplatinum Pt3Fe |
1.7 |
|
Electrum (AuAg) |
3.3 |
|
Sperrylite (PtAs2) |
6.0 |
|
Laurite (RuS2) |
5.2 |
|
Pt-Pd Tellurides |
2.6 |
|
Pt-Pd Sulphides |
80.9 |
Table 3 - Tenor of
PGE in Base Metal Sulphides at Rustenburg
(After Kinloch, 1982)
|
Base Metal Sulphide |
Pt ppm |
Pd ppm |
Rh ppm |
|
Pentlandite |
11 |
140 |
23 |
|
Pyrrhotite |
11 |
0 |
3 |
|
Pyrite |
15 |
34 |
9 |
Other work,
which focussed on the metallurgical implications of the modes of occurrence of
PGE amongst the entire PGE host mineral set, found that silicates and oxides
also accounted for a small amount of PGE in the Merensky [Peyerl, 1983]. Although the purpose of that investigation
was to account for the flotation behaviour of the various PGE hosts, it also
became clear that some 5-10% of all PGE assayed in this ore was present as
solid solution in silicates and oxides.
The degree of enrichment of this silicate/oxide phase was dependent on
the extent of serpentinisation in the particular area of the Bushveld.
When the
relative grain sizes are considered, the minerals hosting PGE in Merensky
become vulnerable to certain evaluation problems. The average grain size of discrete PGM in the
Rustenburg Merensky is 26 mm, with a size
range from 1 to 840 mm.
The fire-assaying of ore and mineral process
samples of this orebody therefore faces a complex task in addressing diverse
mineral hosts across several orders of magnitude in PGE grade.
Technical
Exercises on Replicate Sampling and Assaying
Method
Representative
samples of Merensky mill feed, final tailings and
final concentrate were taken from the Frank Concentrator of Rustenburg Platinum
Mines Limited. The rules for
representative sampling were observed [Lotter,
1995b]. The lead-collection fire-assay
method described in this paper was used throughout the
analysis of the samples. This included
both LT and HT cupellation. Gravimetric
determination was performed on the LT prills.
Mill
Feed
The mill feed
sample was dried in a thermostatically-controlled oven
at 1060 C, and then crushed to a limiting size of 3mm. The crushed product was blended in a spinning
riffler using a specialised technique called ‘odds-and-evens blending’ [Lotter,
1995a]. The spinning riffler is known as
a superior technology for this purpose [Allen, 1990], and is used without
exception where thorough blending is required.
The blended mill feed sample was then subsampled using the spinning
riffler to produce replicate 2.5 kg test lots of ore. The proportionality of the spinning riffler
allowed 25-kg charges, which were batch-weighed, to produce 2.5 kg subsample
units. Test lots of 2.5 kg sample were
chosen randomly from this suite and milled in a laboratory scale ball mill to
produce milled charges at grinds of 40, 50, 60 and 70% mass passing 75 m. The process was wet, i.e. operated at a pulp density similar to that of a primary ball
mill in the production process. This is typically 80% solids w/v. Each milled
charge was quantitatively removed from the mill and filtered
then dried at 106o C.
The milled charges were then separately reblended
and subsampled using the spinning riffler
in preparation for fire-assaying.
This
arrangement is shown in Figure 1.
Figure
1 – Subsampling and Batch Milling Arrangement for
Mill Feed
Final
Concentrate
Similar
arrangements to the mill feed were made, except that
the sample was not crushed and milled.
Drying was performed at a lower temperature (80o C), and
dried final concentrate sample blended using the odds-and-evens method. Smaller subsamples (50 g) were
produced from the spinning riffler.
Three groups
of subsamples were extracted from this set.
1. Group A was selected
and set aside ‘as is’. This group is referred
to as ‘blended’.
2. Group B was prepared
by batch-pulverising 50 g charges with matched masses of analytical silica in a
ring pulveriser. The pulveriser
operation time was 2 minutes, controlled using a timer. This group is referred
to as ‘silica ground’.
3. This pulveriser
was tested for mass balance criteria prior to the testwork
and was found to lose on average 0.2% of the sample mass. The product from this process was called ‘pulverite’.
The reason for producing group B was to test for the effect of
homogenisation on the fire-assay result, since the concentrate as produced by
the mineral process carries PGE in distinctly different mineral hosts that have
natural PGE grades across 5 orders of magnitude. It is believed that
the pulverisation-in-silica process transforms the discrete PGM (such as
ferroplatinum or platinum-palladium sulphides) into an artificial solid
solution in the analytical silica, thus reducing the grade of the ferroplatinum
from some 40% PGE by some three orders of magnitude to an order of several
hundred grams per tonne.
4. Group C was treated
by laboratory scale superpanning to semi-quantitatively remove the discrete PGM
present by gravity separation prior to assaying. A mass balance was determined, and both
gravity concentrate and tailings were assayed. This group is referred
to as ‘superpanned’.
These
arrangements are shown in Figure 2.
Figure 2 – Subsampling and Sample Treatments for Final Concentrate
Final
Tailings
Similar
arrangements to those for mill feed were made, except
that there was no crushing. The tailings
sample, having been produced from a production
grinding process, had a particle size distribution equivalent to 43% mass
passing 75 mm.
Ball-milling of 2.5 kg subsamples of tailings
produced ground charges of 50, 60 and 70% mass passing 75 mm. These milled charges were
filtered and dried as for mill feed.
Figure 3 – Subsampling
and Batch Milling of Tailings
Fire-Assaying
Lead-collection
fire-assaying was used in all cases to estimate the
PGE grade of the processed samples. For
each category of sample, viz. mill
feed, final concentrate and final tailings, the fusion mass of sample, or
analyte mass, was adjusted within the test group so
that prills of comparable mass would be obtained. For example, the final concentrate samples
used 5 g for Group A and 10 g for Group B, so producing similar prill masses. Mill feed analyte mass was 50 g, and final
tailings, 125 g. Within each sample
type, replicate fusions were arranged so that samples
from each grinding condition were present in the 30-fusion tray. In this way, variation in furnace conditions could not be blamed for any difference in the results. Natural standards were included in the fusion
tray layout to cross-check the unknown fusions. The degree of inspection was in the order of 4 standard fusions per tray.
RESULTS
Mill
Feed
The overall
results of the replicate fire-assaying of the mill
feed samples by grind group are summarised in Table 5 and Figure 4. The screened data, i.e. those accepted within the sample mean " 2 sample standard deviations, are
also shown.
Table 4 - Analysis of
Mill Feed PGE Grade by Grind Group
|
Parameter |
Test Condition :
Grind, % Sample Mass Passing 75 Microns |
|||||||
|
40 Raw |
40 Screened |
50 Raw |
50 Screened |
60 Raw |
60 Screened |
70 Raw |
70 Screened |
|
|
Mean Grade g/t PGE |
1.000 |
0.988 |
0.998 |
1.003 |
1.026 |
1.010 |
1.017 |
1.029 |
|
RSD % |
20.0 |
13.3 |
15.2 |
10.9 |
26.0 |
8.8 |
15.6 |
10.9 |
|
n |
143 |
137 |
14.3 |
136 |
107 |
105 |
108 |
103 |
Figure
4 – Effect of Mill Feed Sample Grind on Mean Grade of PGE Estimated
Final
Concentrate
The overall results
of the replicate fire-assaying of the final
concentrate samples by grind group are summarised in Table 5 and Figure 5.
Table 5 - Analysis of
Final Concentrate PGE Grade by Sample Preparation Method
|
Raw Data Parameter |
Test Condition :
Sample Preparation Method |
|||||
|
Group A As-Is |
Group B Pulverised in Silica |
Group C Prior Separation of
Discrete PGM |
||||
|
Raw Data |
Screened |
Raw Data |
Screened |
Raw Data |
Screened |
|
|
Mean Grade g/t PGE |
1.000 |
1.004 |
0.981 |
0.980 |
0.973 |
0.978 |
|
RSD % |
3.06 |
2.28 |
2.66 |
2.45 |
2.14 |
1.45 |
|
n |
60 |
58 |
56 |
54 |
12 |
11 |
Figure
5– Effect of Sample Preparation Method of Concentrate on Mean Grade of PGE
Estimated
Final
Tailings
The overall results
of the replicate fire-assaying of the mill feed
samples by grind group are summarised in Table 6 and in Figure 3. The screened data, i.e. those accepted within
the sample mean " 2 sample
standard deviations, are also shown.
Table 7 - Analysis of
Tailings PGE Grade by Grind Group
Raw Data
|
Parameter |
Test Condition :
Grind, % Sample Mass passing 75 Microns |
|||||||
|
40 Raw |
40 Screened |
50 Raw |
50 Screened |
60 Raw |
60 Screened |
70 Raw |
70 Screened |
|
|
Mean Grade g/t PGE |
1.000 |
1.036 |
1.051 |
1.069 |
1.106 |
1.147 |
1.129 |
1.149 |
|
RSD % |
20.4 |
14.2 |
14.7 |
9.3 |
20.0 |
12.9 |
18.8 |
11.4 |
|
n |
168 |
158 |
135 |
127 |
103 |
97 |
135 |
126 |
Figure 3 – Effect of Tailings Sample Grind on Mean Grade of
PGE Estimated
Tests of Significance
In order to
determine the significance of any observed differences in grade of PGE
estimated by sample treatment group, the Analysis of Variance (ANOVA) was used
[Box et al., 1978]. This process
compares the variance arising within treatment type with variance between
treatment type.
The result is referred to a value of F, the
variance ratio.
Although
similar patterns are seen between raw and screened
data in the above results, the ANOVA was performed on the screened data set
where effects are sharper.
Mill Feed
Table
8 - ANOVA Table
Mill Feed : Screened Data
|
Sum of Squares |
Degrees of Freedom |
Mean Square |
|
|
Between Treatments |
2.46 |
3 |
0.82 |
|
Within Treatments |
147.39 |
477 |
0.31 |
|
Total |
149.86 |
480 |
0.31 |
From these calculations the F ratio for the test of significance is 0.82/0.31=2.65. The critical value of F at the 10% level is approximately 2.1, and at the 5% level, approximately 2.6, so it may be concluded that the grinding treatment of the mill feed produces significantly different PGE grades at the 5% significance level.
Final Concentrate
Table 9 - ANOVA Table
Concentrate :
Screened Data
|
Sum
of Squares |
Degrees
of Freedom |
Mean
Square |
|
|
Between
Treatments |
8511.5 |
2 |
4255.8 |
|
Within
Treatments |
35063.6 |
123 |
285.1 |
|
Total |
43575.1 |
125 |
348.6 |
From these
calculations the F ratio for the test of significance is 4255.8/285.1 =
14.92. The critical value for F at the
0.1% level is approximately 7.32, so the screened are significant at the 0.1%
significance level.
Final Tailings
Table 10 - ANOVA Table
Tailings: Screened Data
|
Sum
of Squares |
Degrees
of Freedom |
Mean
Square |
|
|
Between
Treatments |
0.400 |
3 |
0.133 |
|
Within
Treatments |
9.498 |
505 |
0.019 |
|
Total |
9.898 |
508 |
0.152 |
From these
calculations the F ratio for the test of significance is 0.133/0.019 = 7.08.
The critical values of F for these test conditions are
approximately 5.6, at the 0.1% level, so it may be concluded that the degree of
grind of the tailings sample prior to fire-assaying
significantly affects the grade of PGE estimated.
Proposed
Mechanisms
Mill Feed and
Tailings
Both mill feed and tailings samples show optimised
PGE evaluation at the same grind, viz. 70% mass
passing 75 microns. The effect is clearer
for the tailings, probably because most of the recoverable PGE mineral hosts have been removed by flotation into the final concentrate.
It is suggested that this trend is
consistent with the degree of liberation of the PGE-bearing minerals. The primary lead fusion stage of the
fire-assay process is time- and temperature dependent, inferring that the
kinetics of the reaction are sensitive to the particle size distribution and
the degree of liberation of the minerals of interest. Inadequate liberation in mill feed and
tailings samples would lead to incomplete recovery of the PGE into the lead
button.
Final
Concentrate
The difference in
concentrate grade arising from prior pulverisation
in clean silica, compared to analysis of the concentrate on an ‘as is’
basis, may have arisen from the
heterogeneous nature of this sample material,
i.e. that it carries PGE mineral hosts of varying grades and densities
across several orders of magnitude. It
is believed that the pulverisation process in
transforms these PGE minerals into a more homogenous form with a more
symmetrical disrtribution of replicate values.
A cross-check on which of these two forms of sample
preparation may be found in the result of Group C of this concentrate
experiment. Here the discrete PGM, which
are a high-density mineral group grading in the order of 40-50% PGE, were quantitatively removed using a laboratory scale superpanner prior to the fire-assaying process. This would result in the statistical
stratification of the sample [Cochran, 1947] and a more reliable estimation of
the mean grade of that concentrate sample.
Inspection of the results of Group C shows that the mean
grade estimated is actually in closer agreement with the mean result of group B
than Group A, suggesting that whether the concentrate sample is physically
stratified or physically homogenised, a lower grade of PGE is estimated. This suggests that the result from Group A is
the outlier and is a biased estimate.
Likely Biases
In collating these effects as they would
appear to a production plant’s metal accounting, it is necessary not only to
examine the combined effects of robust mean grades, such as obtained in these
exercises by large data sets, but also to examine the likely biases in
estimating the mean grades of these samples from the small data sets which are
typically used in daily production assaying. This is because skew
and multimodal distributions are prone to misleading estimates of the mean when
estimated from non-robust data sets.
Accepting in the first case that the estimates of the mean
grades are more reliable after outlier removal at the 95% acceptance level, the following
likely biases would develop, using typical plant grinding conditions and
compared to the optimum state of sample preparation:
Table 11
Likely Biases Using Robust Screened Data
|
Sample |
Grade at Production
Grind |
Grade with Optimised Sample Preparation |
Difference % (Production Sample
=100%) |
|
Mill
Feed |
0.988 |
1.029 |
-
4.14 |
|
Concentrate |
1.004 |
0.980 |
+
2.29 |
|
Tailings |
1.051 |
1.149 |
-
9.30 |
This analysis indicates that unless optimised
sample preparation is used, mill feed grade will be understated by some 4.1%;
tailings by some 9.3%, and concentrate overstated by some 2.3%. (A convention of negatively signed bias is used here to indicate underestimation of grade). When combined in a typical mass balance
across the mineral process, these errors would amount to an overstatement of
PGE recovery across flotation processing by some 5.2%, using typical mass
balance data. Put differently, if PGE metal
accounting were to be performed on a reconciled head basis, and a call factor
calculated from the assayed head, an unaccounted gain of some 5.2% would be the
typical result.
Accepting in the second case that the estimates of the means
are robust to the presence of outliers, the following likely biases would
develop, using typical plant grinding conditions and compared to the optimum
state of sample preparation:
Table 12
Likely Biases Using Raw Data with Outliers Present
|
Sample |
Grade at Production
Grind |
Grade with Optimised Sample Preparation |
Difference % (Production Sample
=100%) |
|
Mill
Feed |
1.000 |
1.017 |
-
1.7 |
|
Concentrate |
1.000 |
0.981 |
+
1.29 |
|
Tailings |
1.000 |
1.129 |
-
12.9 |
This analysis indicates that unless optimised
sample preparation is used, mill feed grade will be understated by some 1.7%;
tailings by some 12.9%, and concentrate overstated by some 1.29%. (A convention of negatively signed bias is used here to indicate underestimation of grade). When combined in a typical mass balance
across the mineral process, these errors would amount to an overstatement of
PGE recovery across flotation processing by some 2.5%, using typical mass
balance data. Put differently, if PGE metal
accounting were to be performed on a reconciled head basis, and a call factor
calculated from the assayed head, an unaccounted gain of some 2.5% would be the
typical result.
Geostatistical Considerations
The pattern of these biases appears to be grade-dependent, i.e. lower grades underestimated, and
higher grades over-estimated. Earlier
work in this regard, addressing the bias errors of gold evaluation in the
Conclusions
The investigation of evaluation bias errors of PGE in
Rustenburg Section Merensky ore has shown that:
1. Lead collection fire-assaying,
as practiced in the work performed for this paper, is not robust to the diverse
mineralogical forms in which the PGE occur.
2. Mill feed and tailings samples show
understatement biases of approximately 4.1% and 9.8% respectively. It is believed that
the mechanism is by means of incomplete fusion of the incompletely liberated
PGE minerals, thus the reason why finer grinding of these samples produces
higher mean PGE grade estimates.
3. Concentrate sample material studied
shows an overstatement bias of approximately 2.3%. It is believed that this mechanism is by
means of the heterogeneous nature of the concentrate, viz, carrying PGE
minerals across several orders of magnitude in grade, and of a range of PGE mineral densities. The pulverisation-in-silica
method is likely to have homogenised this arrangement
into an artificial set of minerals with less spread of specific grade.
4. By constructing technical exercises as
described in this paper, the investigator is in a position to minimise these bias errors and obtain more
reliable estimates of grade and recovery. It is possible that other parts of the Bushveld Complex would display similar bias effects, but
with bias coefficients typical to the specific geological area in
question. Wider ranges of grinding
conditions should also be tested.
5. The overall effect of the bias errors
documented for this section of the Bushveld Merensky is in the order of 5.2% expressed as either
recovery or PGE call factor error.
Acknowledgements
Mr BR Beamish, Operations Director, AMPLATS, is due thanks for
permission to publish this paper. Dr LA Cramer, Chief Consulting Metallurgist, Mr GM Wright, Chief Consulting Metallurgist of AMPLATS,
Johannesburg, South Africa, are due thanks for their ongoing encouragement for
this work. They have constantly shown
the leadership necessary in a changing environment to champion the development
of new ideas and fundamental work. My
colleagues HC Munro and RM Williams, with whom the author worked during several
years at Rustenburg in joint technical studies on this subject, provided much
encouragement and advice during the development of the ideas in this paper.
1.
Falconbridge Technology Centre, Falconbridge
Limited,
2.
Minerals Processing
Research Unit,
3.
Geostokos Limited,
References
Adamson, RJ, 1972,
Gold Metallurgy in South Africa, publ.
S. Afr. Chamber of Mines, 1972, pp.178-202.
Allen,
T, 1990, Particle Size
Measurement, 4th
ed., 1990, chap. 1, pp.38-39.
Box,
GEP, Hunter, WG, and Hunter, JS, Statistics for Experimenters, publ. Wiley, 1978, chap. 6.
Cochran, WG, 1946, Relative Accuracy of Systematic and Stratified
Random Samples for a Certain Class of Populations, Ann. Math. Statistics, 1946, 17, pp. 164-177.
Coxon,
CH, and Sichel, HS, 1959, Quality Control of Routine Mine
Assaying and its Influence on Mine Valuation, J. S. Afr. Inst. Min. Metall., 59, pp.489-517.
Eissler, M, 1896,
The Metallurgy of Gold, A Practical Treatise on the Metallurgical Treatment
of Gold-Bearing Ores, publ. Juta, 1896, 4th ed.,pp. 533-570.
Gy, P, 1979, Sampling of Particulate Materials : Theory and Practice,
Elsevier
(publ.), 1979, chap. 28.
Kinloch,
ED, 1982, Regional Trends in the Platinum-Group Mineralogy of the
Critical Zone of the Bushveld Complex,
Krige, DG, 1981, Geostatistics I : Lognormal-de Wijsian Geostatistics for ore Evaluation,
SAIMM Monograph, 1981, pp.1-22.
Krige,
DG, 1962, Statistical Applications of Mine Valuation,
J.Inst. Min. Surv. S. Afr.,
12, 2, pp. 95-136.
Lenehan,
WC, and de L Murray-Smith,
1986,
Assay and Analytical Practice in the South African Mining Industry, publ. S.Afr. Instn. Min. Metall, 1986, chap. V, pp. 41-62.
Lotter,
NO, 1995a, A Quality
Control Model for the Development of High-Confidence Flotation Test Data,
unpubl. M.Sc.(
Lotter, NO, 1995b, Review
of Evaluation Models for the Representative Sampling of
Munro, HC, 1993, Personal
Communication, Rustenburg, August 1993.
Peyerl, W, 1982, The
Metallurgical Implications of the Mode of Occurrence of Platinum-Group Metals
in the Merensky Reef and UG-2 Chromitite of the Bushveld Complex, Spec.
Publ. Geol. Soc. S. Afr., (7), 1983, pp. 295-300.
Robert, RVD, van Wyk,
E, and Steele, TW, 1975, The Efficiency of Various Matrix
Elements on the Efficiency of the Fire Assay Procedure Using Nickel Sulphide as
the Collector, National
Institute of Metallurgy, Report No 1705, Johannesburg, South Africa, 1975.
Sichel, HS, 1947, An Experimental and Theoretical
Investigation of Bias Error in Mine Sampling with Special reference to
Narrow Gold Reefs, Trans. Inst. Min. Metall.(
Viljoen,
MS, and Hieber, RW, 1982, The Rustenburg Section of Rustenburg
Platinum Mines Limited, with Reference to the Merensky Reef, Mineral deposits
of South Africa, Annhauser, CR, and Maske, S, (eds) Publ. Geol. Soc. S. Afr.,
1982, vol 2, pp.1107-1135.
Williams, RM, 1994, Personal
Communication, Rustenburg,
June 1994.
[Manuscript revised