Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

Optimisation of the grade control procedures

at the Yandi iron-ore mine, Western Australia:


geostatistical approach
M. Z. Abzalov*1, B. Menzel2, M. Wlasenko2 and J. Phillips2
Effectiveness of ore grade control at operating mines depends on both the quality and quantity of
the samples used. Therefore, optimisation of grade control procedures requires analysis of sample
quality and their spatial distribution. This approach, implying quantitative estimation of both these
factors and quantification of their contribution to the grade control errors, was used for comparing
two different grade control procedures at the Yandicoogina (Yandi) iron-ore open pit mine, located
in the eastern part of the Pilbara region of Western Australia. At Yandi, pisolitic iron oxide
mineralisation is distributed within a meandering palaeoriver channel, characterised by abundant
clay pods that contain the deleterious components; in particular Al2O3 and SiO2. The irregular
shapes and highly erratic distribution of the clay pods are a serious challenge for grade control at
this mine. Current grade control procedures at the Yandi open pit are based on sampling 10 m
deep blast holes on a 666 m grid. An alternative approach was considered of using reverse
circulation (RC) holes. However, study of the sample duplicates and direct comparison of the blast
holes and RC holes drilled at distances approximately 1 m apart (i.e. twin holes) have revealed that
use of RC drilling does not guarantee an improvement in the quality of the samples. Reverse
circulation samples exhibit precision errors similar to that of the blast holes and both sets of data are
characterised by large nugget effects on the variograms of Al2O3 values. It was also noted that RC
samples were biased, underestimating Al2O3 and SiO2 grades, and overestimating Fe grades. The
Sequential Gaussian conditional simulation model was used for estimation of grade control errors
associated with both grade control techniques. This model has shown that using a grid of 25625 m
for grade control leads to 12?3% of the selective mining unit sized blocks being misclassified, which
is more than a two-fold increase in comparison with current procedures based on sampling blast
holes, when approximately 5?8% of SMU blocks are misclassified. In general, comparative analysis
of the two grade control approaches at the Yandi mine shows that using the RC holes drilled as the
25625 m grid will not reduce the grade control errors. In fact, it can produce the opposite effect,
significantly increasing the number of misclassified ore and waste blocks.
Keywords: Iron, Grade control, QAQC, Conditional simulation, Yandi mine, Pilbara

This paper is part of a special issue on grade control and geostatistics of iron ore deposits

Introduction and for determining the destination of the material types


depending on the ore characteristics.
Grade control data are acquired at the pre-production Choice of the grade control procedures requires
stage to assure that ore that is being hauled from the finding the optimal combination of quality and quantity
open pit meets the required grade and metallurgical of grade control data to allow cost effective delineation
specifications. The detailed information collected at the of ore and waste blocks and accurate determination of
grade control stage is used for separating ore from the characteristics of the mined ore. In practice, selection
waste, usually referred to as ‘ore-waste classification’, of the grade control techniques is often based on
personal preferences of the mine managers and is rarely
supported by quantitative estimates of the potential
1
Rio Tinto Exploration, 37 Belmont Avenue, Belment, WA 6104, Australia grade control errors associated with different data
2
Rio Tinto Iron Ore, Resource delelopment, Perth, WA 6000, Australia qualities and the spatial distribution of the samples.
*Corresponding author, email marat.abzalov@riotinto.com An alternative approach to optimisation of grade

ß 2010 The Australasian Institute of Mining and Metallurgy


Published by Maney on behalf of the Institute and The AusIMM
Received 19 January 2011; accepted 3 May 2011
132 DOI 10.1179/1743275811Y.0000000007 Applied Earth Science (Trans. Inst. Min. Metall. B) 2010 VOL 119 NO 3
Abzalov et al. Grade control at Yandi mine, Pilbara region, Australia

1 Map showing location of the Yandicoogina (Yandi) mine in Western Australia

control procedures and parameters is Conditional structure that is the course of an ancient river. It is also
Simulation (Goovaerts, 1997; Chilès and Delfiner, known as a pisolitic iron-ore deposit, because of the
1999), the geostatistical technique which allows the nature and geometry of the iron oxides that are infilling
generation of mathematical models of the ore bodies the palaeochannel (Ramanaidou et al., 2003). CID-type
and the use of these models for estimation the grade is the preferred terminology for describing this type of
control errors associated with the different proposed deposit in the Pilbara region (Ramanaidou et al., 2003).
drill grids (Abzalov and Bowers, 2009). The palaeochannel is approximately 80 km long and
The geostatistical approach was applied at the 600 m wide (Fig. 1) and marks the course of mid-
Yandicoogina (Yandi) iron-ore open pit mine (Fig. 1). Miocene creeks that cut into the weathered surface of
At this mine the Al2O3 and SiO2 are the main deleterious the Weeli-Wolli Formation. The shape of the palaeo-
components and together with Fe grade are used for channel is strongly meandering in plan (Figs. 1 and 2),
classification of the mined materials. In particular, but relatively simple and saucer-shaped in cross-section
accurate knowledge of the Al2O3 and SiO2 contents is with no significant folds or faults (Fig. 3).
important for separation of the mined ore from waste Sedimentary formations that make up the palaeo-
and for mine production planning, and therefore their channel have been subdivided into several units (Fig. 3)
content needs to be accurately estimated during grade which are overlain by barren clay-rich sediments
control. representing a transported regolith and thin alluvial
Current grade control procedures at the Yandi open cover. All sedimentary formations within the palaeo-
pit are based on sampling 10 m deep blast holes channel contain iron oxides, however ore grade miner-
distributed on 666 m grid. An alternative approach alisation is distributed within a relatively thin zone,
was considered for grade control based on sampling of known as the Main Ore Zone (Table 1); this zone is
reverse circulation (RC) holes, drilled at the pre- approximately 50 m in thickness in the centre of the
production stage. The drill spacing between RC holes channel, thinning toward the channel margins (Fig. 3),
is 25625 m, i.e. significantly broader than the currently and is distributed as a continuous bed along the strike of
used blast holes. One of the reasons for the proposal to the palaeochannel (Fig. 2).
change grade control procedures from blast holes to RC The iron-oxide pisoliths filling the palaeochannel
holes was an assumption that RC samples are of better generally have hematitic cores surrounded by goethitic
quality than samples collected from the blast-hole cones. rims and cemented by a goethitic matrix to form an
It was important to check this assumption before any indurated rock. Composition of the iron-oxide miner-
decision was made. The consequences of the four times alisation and ore quality are relatively consistent except
increased distance between the grade control samples for the distribution of Al2O3 (Fig. 2) and to a lesser
also had to be quantitatively estimated before mine degree SiO2, i.e. the main deleterious components of the
management could make an informed decision on the Yandi ore. The average coefficient of variation of Al2O3
two alternative approaches. Methodology of the study ranges from 93?2 to 102?6% in the Main Ore Zone,
and findings are described in this paper. which is the highest variability among all measured
components of the iron ore at Yandi (Table 1). The level
of Al2O3 and SiO2 content is lower and less variable at
Geological background the centre of the channel than at the edges (Fig. 2).
The Yandi open pit mine is located in the eastern Pilbara Distribution of these components, in particular Al2O3,
region of Western Australia (Fig. 1). It hosts pisolitic reflects the distribution of clay minerals (e.g. kaolinite –
iron-oxide mineralisation distributed within part of the Al2(OH)4[Si2O5]) which are more abundant at the edges
Marillana – Yandicoogina – Weeli Wolli palaeoriver of the palaeochannel (Figs. 2 and 3) and toward its base.
channel (Hall and Kneeshaw, 1990, Ramanaidou et al., Systematic increase in the average Al2O3 content is
2003). The deposit is usually referred to as Channel Iron shown in Fig. 2 from GVU though GVL (both parts of
Deposit (CID-type), emphasising the nature of the host the Main Ore Zone, see Fig. 2) to LGC and basal

Applied Earth Science (Trans. Inst. Min. Metall. B) 2010 VOL 119 NO 3 133
134
Abzalov et al.

Applied Earth Science (Trans. Inst. Min. Metall. B)


location of the cross-section shown in Fig. 3

2010
Grade control at Yandi mine, Pilbara region, Australia

VOL
average content in GVU is higher than in GVL
conglomerates (Table 1) and is consistent with increase
Al2O3 (wt-%). The map was generated during the

rock faces exposed in the Yandi open pit (Fig. 4b). SiO2
in the clay content with depth, as can be observed at the
2 Maps of the mine benches showing distribution of the

also increases at the channel margins; however, its


holes and the blast-hole assays (Fig. 5). Location of
the map area is shown in Fig. 1. Line ‘A-B’ denotes
course of this study using available diamond/RC drill

119
3 NO
Table 1 Distribution of the main iron ore components between stratigraphic units composing the Yandi paleochannel and the basement rocks*

Stratigraphic Fe% Al2O3% SiO2% P% LOI%


unit (explanation Number
in Fig. 3) of samples Mean Range SD CV% Mean Range SD CV% Mean Range SD CV% Mean Range SD CV% Mean Range SD CV%

ALL 3048 29.08 8.68–61.21 7.12 24.5 6.5 1.37–24.81 3.4 52.3 45.51 6.27–79.41 13.14 28.9 0.027 0.005–0.075 0.006 22.2 4.19 1.54–24.00 1.85 44.2
ECC 2425 33.9 2.93–59.70 11.07 32.7 14.72 0.72–32.30 4.55 30.9 25.36 1.60–62.83 11.14 43.9 0.017 0.003–0.130 0.009 52.9 9.08 3.10–18.07 1.66 18.3

WCH 2153 50.08 2.16–63.26 8.38 16.7 6.79 0.41–30.16 4.37 64.4 11.64 0.02–55.02 6.96 59.8 0.038 0.004–0.169 0.015 39.5 8.71 4.02–20.21 1.5 17.2
GVU Main 14 921 59?04 4?60–64?54 2?55 4?3 1?34 0?18–29?99 1?25 93?3 4?98 1?95–53?55 2?37 47?6 0?044 0?007–0?300 0?011 25?0 8?94 2?51–13?98 1?13 12?6
ore zone
GVL 16 558 58?17 0?20–62?26 3?12 5?4 1?56 0?24–31?19 1?6 102?6 3?85 1?61–76?89 2?84 73?8 0?062 0?010–0?197 0?017 27?4 11?05 2?79–13?84 0?86 7?8
LGC 4118 50.44 1.75–60.79 9.18 18.2 4.89 0.17–31.17 4.17 85.3 10.73 2.01–78.37 9.49 88.4 0.076 0.006–0.750 0.03 39.5 11.44 3.69–14.30 1.07 9.4

BCC 2449 45.83 1.88–64.31 10.68 23.3 8.69 0.47–35.60 5.14 59.1 15.1 1.53–88.26 13.34 88.3 0.064 0.004–0.311 0.034 53.1 9.32 1.87–20.38 2.07 22.2
WW 574 39.37 0.79–63.20 14.23 36.1 9.04 0.38–33.98 6.76 74.8 24.84 1.15–88.43 19.45 78.3 0.06 0.004–0.397 0.042 70.0 8.77 1.35–16.95 3.06 34.9
*SD5standard deviation; CV% (coefficient of variation)5100%6(SD/mean).
Abzalov et al. Grade control at Yandi mine, Pilbara region, Australia

3 Cross-section of the Yandi palaeochannel generalised from the mine geology documentation at the Yandi pit. Clay-rich
areas are delineated from the reserve model. Location is shown in Fig. 2

(Table 1), mainly because of higher content of opaline samples with composites of full blast-hole cones. In
silica infilling pores in the pisolitic iron ore. total, six blast-hole cones have been processed.
Clay minerals at the Yandi deposit occur as irregular N For diagnostics of possible biases, twin-hole pairs,
disseminations (Fig. 4a), coating joint surfaces and as where distances between RC and blast holes are
clay pods of highly irregular shapes, including lenses, approximately 1 metre, have been analysed. The
desiccation joint infillings (Fig. 4b) and void-fills varying methodology of the twin holes analysis is explained
from simple rounded or angular occurrences to complex in Abzalov (2009).
amoeba-like forms (Fig. 4b and c). Clay pods vary in
size from centimetres to several metres across (Fig. 4).
N Precision (repeatability) of the analytical data was
estimated by analysing duplicate samples. Only field
Accumulation of clay in the palaeochannel initially duplicates (Abzalov, 2008) have been used in this
occurred during deposition of the iron-oxide pisoliths and study. Precision error is estimated and reported as the
continued during the diagenetic cementation and super- average coefficient of variation (CV%) of the paired
gene weathering of consolidated iron-oxide rich sediments data (equation (1)). This estimator is useful for
infilling the channel. During weathering, the clay was variables characterised by strongly skewed statistical
introduced from the weathered basement rocks and from distributions and where outliers are present. It has
clay-rich sediments overlying the palaeochannel (Fig. 4c). been proposed as the universal measure of relative
Prolonged exposure to hydration and oxidation processes precision error in mine geology applications (Stanley
below the ancient water table has enhanced remobilisation and Lawie, 2007; Abzalov, 2008)
of the clay minerals into voids, creating the final ( " #)1=2
distribution pattern of Al2O3 and SiO2 at the Yandi mine. 2X N
(ai {bi )2
CV%~100| (1)
N i~1 (ai zbi )2
Study methodology and data
In the current study 553 blast-hole sample duplicate
Comparative analysis of the blast holes and the RC pairs and 1040 RC and diamond sample duplicate
drilling has been performed in this study to determine the pairs have been analysed.
optimal grade control approach at the Yandi mine. In
particular, the potential errors which the two approaches
N The variance of duplicate samples has been
compared with the nugget effect [Nugget effect is
produce in classification of the mined materials as ore and a geostatistical term defining ‘...an apparent
waste had to be quantitatively estimated and compared. discontinuity in the experimental variogram near
The sources of potential grade control errors are sample the origin caused by measure-ment errors or by
quality (accuracy and precision) [Accuracy is a measure nested structures that have ranges smaller than the
of how close the sample value is to the true grade; and sampling interval, or both.’ (Olea, 1991, p. 54)] of
precision, or repeatability, is a measure of how close the pair-wise relative variograms. Conventional
sample values are to one another (Abzalov, 2011).] and formula of the pair-wise relative variogram
their spatial distribution, in particular the distance (Goovaerts, 1997) is as follows (equation (2))
between samples. The study was mainly based on data
collected from bench 500–510 m RL of the Yandi open 2X N
½Z(xi ){Z(xi zh)2
pit mine (Fig. 5), an area which contains both blast-holes cPWR (h)~ (2)
N i~1 ½Z(xi )zZ(xi zh)2
and resource definition drill holes (diamond and RC).
The drill-hole samples were composited to the same where Z(x) is a value of variable Z at the location
bench height as the blast holes. In total, bench 500–510 m x, and h is a vector separating Z(x) from Z(xzh).
RL contains 35137 blast-hole samples and 921 compos- N CV% (equation (1)) is a square root of equality 2
ited drill-hole samples, of which 4326 blast-hole samples and it represents a case when vector h separating
and 68 drill-hole composites are located in the detailed the data pairs is equal to zero.
study area (Fig. 5). In order to quantify the consequences of the four times
Quality of the grade control samples has been increased distance between samples if RC drilling ra-
quantified by estimating their accuracy and precision. ther than blast-hole drillings is implemented for grade
N Accuracy has been estimated by comparing blast- control purposes, the number of mining blocks misclassi-
hole samples and bench-height composited RC fied by both approaches has been assessed. The estimation

Applied Earth Science (Trans. Inst. Min. Metall. B) 2010 VOL 119 NO 3 135
Abzalov et al. Grade control at Yandi mine, Pilbara region, Australia

5 Map showing location of the blast-hole samples and


the drill holes used for resource definition (red dots).
Detailed study area is the area for which the SGS mod-
els were generated

within the detailed study area (Fig. 5). The


simulation used a sequential neighbourhood, with
a dilation radius of 400 m (X)6200 m (Y)65 m
(Z). The maximum number of the conditioning data
nodes and the simulated nodes were both set to 40.
Experimental variograms and their models have
been published elsewhere (Abzalov et al., 2007). As
the SGS method assumes a multi-Gaussian property
of the studied random variable and its diffusive
distribution model, these assumptions need to be
tested prior to application of the modelling meth-
odology. In the current study, the border effect
was tested by calculating the ratios between cross-
variograms of the indicators and indicator variograms
(Abzalov and Humphreys, 2002). Multi-Gaussianity
was tested by determining variograms of indicators
calculated for the chosen data percentiles and com-
paring them with indicator variograms calculated for
the same percentiles of the Gaussian transformed data
4 Distribution of the clay minerals at the Yandi mine: a (Goovaerts, 1997);
mixture (CL) of clay and limonite filling voids and
replacing iron-oxide pisoliths; b clay pods (CL) irregu-
N representative realisation of the simulated data, whose
grades and spatial distribution patterns closely repro-
larly distributed in the iron-ore bed exposed at the pit duce that of the blast-hole samples, have been chosen
wall; c clay minerals (white colour) filling the joints in for further study. This exhaustive set of data, distri-
the weathered iron-stone unit (dark brown) buted as a 161 m grid, was used as a reference model
for estimating the grade control errors, i.e. this model
is considered as the ‘True’ iron-ore mineralisation;
was based on the geostatistical models generated using se-
quential Gaussian conditional simulation (SGS) technique
N minimum size of the block that can be selectively
mined at the Yandi open pit is 25625610 m. The
(Goovaerts, 1997; Chilès and Delfiner, 1999) implemented cut-off grade for Al2O3 is taken as ,1?5 wt-%. Thus,
in commercial geostatistical software Isatis (Bleines et al., the selectively minable unit (SMU) is considered as ore
2001). Sequential Gaussian conditional simulation is a if the estimated Al2O3 grade is less than 1?5%. For
Gaussian-based method of conditional simulation which each (SMU) their ‘True’ contents of Al2O3 and SiO2
uses data transformed to a Gaussian distribution with a have been estimated using exhaustive data, distri-
zero mean and a unit variance (i.e. Gaussian anamorpho- buted as the 161 m grid, and representing the ‘True’
sis) and applies a Monte-Carlo simulation algorithm to the mineralisation;
regionalised variables. One model was generated to
represent the current grade control practice with blast-hole
N the grade control errors in classification of the
SMU size blocks have been estimated for two
samples distributed as a 666 m square grid, and the grids, one of 565 m which is similar to the current
alternative model represents the RC holes distributed as a distribution of the blast holes, and one of
25625 m square grid. The SGS models were generated for 25625 m which was considered as an alternative
the detailed study area shown in Fig. 5. grade control grid. For both of these grids the
Details of SGS technique, as applied in the current metal grades of the SMUs were estimated by the
study, and the parameters used, are as follows: ordinary kriging (OK) technique (Goovaerts,
N blast-hole grades of Al2O3 and SiO2 have been 1997). Errors in classification of the SMU size
simulated to an exhaustively dense (161 m) grid blocks were assessed by comparing their grades

136 Applied Earth Science (Trans. Inst. Min. Metall. B) 2010 VOL 119 NO 3
Abzalov et al. Grade control at Yandi mine, Pilbara region, Australia

6 Comparison of the pairs of sample duplicates. a–c blast-hole data: a Al2O3; b SiO2; c Fe. d–f RC and diamond holes:
d Al2O3; e SiO2; f Fe. RMA denotes the reduced major axis (Sinclair and Bentzen, 1998; Abzalov, 2008). Deviation of
the RMA from 1 : 1 line indicates for a possible accuracy issue between compared pairs of data

estimated by OK with initially generated ‘True’ Results


values. In other words, the estimated OK grades,
representing the grade control models, were Duplicate samples
compared with the initially generated exhaustive Results of the duplicate sample analysis show that both
data set which was considered ‘True’. types of data, blast-hole cuttings and RC samples,

Table 2 Statistical parameters deduced from pairs of the sample duplicates

Yandi mine Hopes Down West Angeles

Blast holes RC/diamond (10 m composites) RC holes RC holes

Al2O3 SiO2 Fe Al2O3 SiO2 Fe Al2O3 SiO2 Fe Al2O3 SiO2 Fe

Number of sample pairs 553 553 553 97 93 93 228 228 228 8088 8088 8088
Mean (1) % 1.4 4.4 58.8 1.6 4.6 58.8 2.1 3.5 62.2 5.7 12.6 51.3
Mean (2) % 1.4 4.4 58.7 1.7 4.8 58.6 2.1 3.5 62.2 5.7 12.5 51.4
Standard deviation (1) 0.8 1.4 1.2 1.3 2.0 2.5 6.6 14.6 14.1
Standard deviation (2) 0.8 1.5 1.3 1.3 2.7 3.3 6.5 14.5 13.9
Correlation 0.88 0.88 0.8 0.99 0.97 0.97 0.92 0.95 0.94 1 1 1
CV% (estimated using equatjion (1)) 14?7 8?8 1?0 11?6 7?4 1?5 12?4 13?7 1?4 6?9 7?0 2?2

Applied Earth Science (Trans. Inst. Min. Metall. B) 2010 VOL 119 NO 3 137
Abzalov et al. Grade control at Yandi mine, Pilbara region, Australia

a map showing location of the studied pairs of holes; b Al2O3; c SiO2; d Fe


7 Comparison of the blast-hole data with resource definition drilling (RC and diamond holes) separated by a distance of
1m

exhibit similar levels of precision errors (Table 2). In and highly repeatable estimation of the Fe grade, with
both cases Al2O3 and SiO2 are characterised by CV% less than 2%. The finding is consistent with the
variability significantly larger than Fe grades. CV% global statistical parameters of the iron-ore components
values estimated from the blast-hole duplicates are at the Yandi deposit (Table 1) and their geological
14?7% for Al2O3 (Fig. 6a), 8?8% for SiO2 (Fig. 6b) and distribution (Figs. 2–4) and clearly shows that using RC
1?0% for Fe (Fig. 6c, Table 2). Duplicates of the RC and hole samples instead of sampling blast-hole cuttings will
diamond samples show similar variation. Their esti- not improve repeatability of the grade control assays.
mated CV% values are 11?6% for Al2O3 (Fig. 6d), 7?4%
Full blast-hole sample cone analysis
for SiO2 (Fig. 6e) and 1?5% for Fe (Fig. 6f, Table 2).
These values are similar to precision errors observed in The results of the comparison of full blast-hole cone
RC samples collected at other mines of the Pilbara grades from six locations with the related blast-hole
region (Abzalov, 2008), in particular at Hope Down, as samples and RC composites are summarised in Table 3.
Blast-hole sample grades of Al2O3 and SiO2 are slightly
shown in Table 2 for comparison. The lower precision
higher than their corresponding values in the full cone
error of the Al2O3 assays at the West Angelas deposit
assays, whereas Fe grades are lower. Comparison of the
(Table 2) is explained by the more even distribution of
RC composites with full cone assays shows a similar
the iron-ore components, in particular the absence of
trend; however, the magnitude of the differences is larger
clay pods at West Angelas.
than that of the blast-hole samples. These results clearly
According to this study both types of samples exhibit
show that RC and blast-hole samples are equally biased
a similar level of repeatability. Their precision errors,
in comparison with full cone assays. Both grade control
estimated from the pairs of sample duplicates, vary sampling techniques underestimate Al2O3 and SiO2 and
approximately from 12 to 15% for Al2O3 and from 9 to overestimate Fe grade. Although the study of only six
13% for SiO2. Both sampling methods produce a reliable full cones is insufficient for conclusive assessment of
magnitude of the sample biases, it indicates that using
RC drilling for grade control at this mine will not
Table 3 Comparison of the average grades of the six full necessarily provide non-biased samples. Data from this
cones with average grades of the corresponding
study also show that RC samples can exhibit larger bias
blast-hole and composited RC samples
that blast-hole samples.
Fe/wt-% SiO2/wt-% Al2O3/wt-%
Twinned holes analysis: short range variability
Full cone 58.18 6.38 1.14 study
Blast-hole samples 58.59 6.16 1.06 The contribution of short range variability (i.e. the grade
RC samples 59.01 5.8 1.02
variation occurring at distances less than the proposed

138 Applied Earth Science (Trans. Inst. Min. Metall. B) 2010 VOL 119 NO 3
Abzalov et al. Grade control at Yandi mine, Pilbara region, Australia

a map showing location of the studied pairs of holes; b Al2O3; c SiO2; d Fe


8 Comparison of the blast-hole data with resource definition drilling (RC and diamond holes) separated by a distance of 10 m

grade control grids) to the global variance of the grade Estimation of the grade control errors using the
control samples has been assessed by comparison of the SGS model
blast-hole grades with grades of closely drilled RC (and/ The SGS model was constructed for Al2O3 and SiO2, the
or diamond) holes. most erratically distributed variables in the Yandi ore.
Results show that blast-hole grades are consistent Grades of the SMU-sized blocks were modelled from the
with RC (and/or diamond) grades when the holes are data, distributed as 565 m and 25625 m grids. Results
drilled as twins (Abzalov, 2009) with an average distance are summarised on the scatter-diagram (Fig. 9) and
between them of approximately 1 m (Fig. 7a–d). In this generalised map of the detailed study area (Fig. 10).
case, the precision errors deduced from the matching Scatter-diagrams (Fig. 9) show the distribution of the
pairs of blast holes and RC (and/or diamond) holes are estimated (kriged) grades of the SMU size blocks plotted
comparable to the errors calculated from duplicates of against their ‘True’ values. The cut-off values are also
the blast-hole samples (Table 2). shown on the diagrams. They divide the plot area into
When the distance between holes increases the varia- four quadrants which are used for estimation the
tion between grades of the matching pairs of holes also number of misclassified blocks (Fig. 9a). In the current
increases. Holes separated by a distance of 10 m show an study, the ‘True’ grade was known a priori from an
excessively poor repeatability of the results (Fig. 8a–d) exhaustive set of the simulated data. Comparison of the
with the CV% value of Al2O3 equals to 35?7% (Fig. 8b). estimated grades of the SMU blocks with their ‘True’
This indicates that significant variations of ore compo- values shows that approximately 5?8% of the SMU
nents occur over relatively short distances. Therefore, blocks are misclassified when grade control sampling is
using a sampling grid of 25625 m will be suboptimal for performed at a 565 m grid (Fig. 9b). The error
grade control purposes at the Yandi mine. increases to 12?3% of misclassified blocks when the
sampling grid increased to 25625 m (Fig. 9c).
Variograms The ‘True’ ore and waste blocks are defined by the
Pair-wise relative variograms have been calculated for Fe, 565 m and 25625 m (Fig. 10) sampling grids. These
Al2O3 and SiO2. The experimental variograms and fitted clearly show that most of the ‘True’ waste blocks could be
models are shown in the Appendix of the paper. Variograms missed and erroneously classified as ore if grade control
of the Gaussian transformed variables, which were used for had been performed using a 25625 m sampling grid.
the SGS model have been published elsewhere (Abzalov et
al., 2007). The variograms of Al2O3 and SiO2 exhibit a local
component, which indicates that significant variability of
Discussion and conclusions
these components occurs at distances less than 20 m. It is well known among mining geology practitioners
Nugget effect of the variograms is in good accordance with that quality of the grade control procedures depends on
precision errors of the Fe, Al2O3 and SiO2 assays estimated both the quality and quantity of the samples. Just
from the sample duplicates (Table 2, Appendix). improving the sample quality does not always lead to

Applied Earth Science (Trans. Inst. Min. Metall. B) 2010 VOL 119 NO 3 139
Abzalov et al. Grade control at Yandi mine, Pilbara region, Australia

a sketch explaining principals of the classification errors calculation using the error quadrants; b Al2O3 grades of SMU
blocks estimated using 565 m data versus ‘True’ grades; c Al2O3 grades of SMU blocks estimated using 25625 m data
versus ‘True’ grades
9 Diagrams of the estimated versus ‘True’ Al2O3 grades of the SMU blocks showing the increased number of misclassi-
fied SMUs when grade control drill spacing increases from 565 m to 25625 m

better defined ore and waste blocks if the spacing chosen Optimisation of grade control procedures should be
for the grade control samples is excessively broad. Thus, based on comparative analysis of the sampling techni-
a robust grade control procedure requires analysis of ques, supported by geostatistical modelling. In particular,
sample quality and their spatial distribution and a the errors associated with different sample spacings
quantitative estimate of the contribution of these factors should be quantitatively estimated using conditional
to the grade control errors. simulation methods. This methodology was used in the
current study to compare two different grade control
procedures for use at the Yandi iron-ore open pit mine.
At Yandi, the short term mine planning and production
quality control require accurate estimation of the metal
contents in SMU approximately 25625610 m in size.
For each of the SMU size blocks it is necessary to
accurately determine the Fe grade and contents of the
deleterious components, in particular Al2O3 and SiO2.
These components, in particular Al2O3, exhibit signifi-
cant variations at short distances which is related to
their presence in clay minerals, mainly distributed as
clay pods that are highly variable in size and shape
(Fig. 4). Given the deleterious nature of these two oxides
and their erratic distribution they represent the main
focus for grade control programs at Yandi.
Current grade control approach is based on sampling
of the blast-hole cuttings. The alternative technique was
to use closely spaced RC holes. The original assumption
was that the sample quality of the RC drilling would be
higher and therefore the use of RC drilling would allow
a better estimation of the grades of the SMU blocks.
Consideration for using RC holes for grade control at
Yandi mine was partially encouraged by successful use
of this technique for grade control purposes at the West
Angelas iron ore mine, which is also located in the
Pilbara region of Western Australia (Harmsworth et al.,
1990). However, at Yandi, study of the sample
duplicates and direct comparison of twin blast holes
and RC holes drilled approximately with 1 m separation
has revealed that the use of RC drilling does not
guarantee an improvement in the quality of the samples.
RC samples exhibit precision errors similar to that of the
blast holes (Table 2; Fig. 6). It was also noted that RC
samples proved to be biased, underestimating Al2O3 and
SiO2 grades, and overestimating Fe grades, when
10 Maps of the ‘Detailed Study Area’ showing distribution compared to full cone assays (Table 3).
of the ORE and WASTE classified using estimated Structures of the variograms calculated using blast-
Al2O3 grades of the SMU blocks hole samples is similar to the variograms calculated from

140 Applied Earth Science (Trans. Inst. Min. Metall. B) 2010 VOL 119 NO 3
Abzalov et al. Grade control at Yandi mine, Pilbara region, Australia

diamond and RC holes (Appendix). In particular, the Abzalov, M. Z. 2009. Use of twinned drill-holes in mineral resource
estimation, Explor. Min. Geol., 18, (1–4), 13–23.
variograms of the Al2O3 values calculated from blast-hole
Abzalov, M. Z. 2011. Sampling errors and control of assay data quality
samples and RC/diamond samples both characterised by in exploration and mining geology, in Application and experience
large nugget effect (Appendix). This is consistent with of quality control, Vienna, InTECH, 611-644. Open on-line
large variability of the Al2O3 contents in the iron ore and access: http//www.intechopen.com/articles/show/title/sampling-
large precision errors deduced from comparison samples errors-and-control-of-assay-data-quality-in-exploration-and-
mining-geology.
and their duplicates (Tables 2 and 4). Abzalov, M. Z. and Bower, J. 2009. Optimisation of the drill grid at the
The SGS model concurs with the results of the Weipa bauxite deposit using conditional simulation, Proc. 7th Int.
twinned holes analysis, showing that significant varia- Min. Geol. Conf., 247–251, Melbourne, The Australasian
tions of the Al2O3 and SiO2 components occurs at Institute of Mining and Metallurgy.
Abzalov, M. Z. and Humphreys, M. 2002. Resource estimation of
distances less than 10 m (Figs. 7 and 8). Comparing structurally complex and discontinuous mineralisation using
SGS models generated using samples distributed as non-linear geostatistics: case study of a mesothermal gold
565 m and 25625 m grids has shown that increasing deposit in northern Canada, Explor. Min. Geol., 11, (1–4), 19–29.
the drill spacing from 565 m (approximating the Abzalov, M. Z., Menzel, B., Wlasenko, M. and Phillips, J. 2007. Grade
control at Yandi iron ore mine, Pilbara region, Western Australia
current blast-hole spacing), to 25625 m (which was – comparative study of the blast-holes and reverse circulation
considered for the RC-based grade control approach) holes sampling, Proc. Iron Ore Conf., 37–43, Melbourne, The
will lead to an approximately two-fold increase in the Australasian Institute of Mining and Metallurgy.
number of misclassified SMU-sized blocks (Figs. 9 and Bleines, C., Deraisme, J., Geffroy, F., Perseval, S., Rambert, F.,
Renard, D. and Touffait, Y. 2001. ISATIS software manual,
10). This generates increased dilution and ore loss, and
Fontainebleau, Geovariances and Ecole des Mines de Paris.
resulting in lower project value. Chilès, J.-P. and Delfiner, P. 1999. Geostatistics: modelling spatial
In general, comparative analysis of the two grade uncertainty, New York, John Wiley & Sons Inc.
control approaches at the Yandi mine has shown that Goovaerts, P. 1997. Geostatistics for natural resources evaluation, New
York, Oxford, Oxford University Press.
using RC holes drilled on a 25625 m grid will not
Hall, G. C. and Kneeshaw, M. 1990. Yandicoogina – Marillana pisoltic
reduce the grade control errors. In fact, it can produce iron deposits, in Geology of the Mineral Deposits of Australia
opposite effect, significantly increasing the number of and Papua New Guinea, (ed. F. E. Hughes), 1581–1586,
misclassified ore and waste blocks. Melbourne, The Australasian Institute of Mining and Metallurgy.
Harmsworth, G. C., Kneeshaw, M., Morris, R. C., Robinson, C. J. and
Shrivastava, P. K. 1990. BIF-derived iron ores of the Hamersley
Acknowledgements Province, in Geology of the Mineral Deposits of Australia and
Papua New Guinea, (ed. F. E. Hughes), 617–642, Melbourne, The
Helpful discussions with colleagues from Rio Tinto Iron Australasian Institute of Mining and Metallurgy.
Ore on the various estimation and risk assessment Olea, R. A. (ed.) 1991. Geostatistical glossary and multi-lingual
techniques are gratefully acknowledged. The authors dictionary, New York, Oxford, Oxford University Press.
also wish to thank G. Broadbent and the anonymous Ramanaidou, E. R., Morris, R. C. and Horwitz, R. C. 2003. Channel
iron deposits of the Hamersley Province, Western Australia, Aust.
reviewers of the AES journal for many useful comments. J. Earth Sci., 50, 669–690.
Stanley, C. R. and Lawie, D., 2007. Average relative error in
References geochemical determinations: clarification, calculation and a plea
for consistency, Explor. Min. Geol., 16, (3–4), 265–274.
Abzalov, M. Z. 2008. Quality control of assay data: a review of Sinclair, A. J. and Bentzen, A. 1998. Evaluation of errors in paired
procedures for measuring and monitoring precision and accuracy, analytical data by a linear model, Explor. Min. Geol., 7, (1–2),
Explor. Min. Geol., 17, (3–4), 1–14. 167–173.

Appendix

11 Experimental pair-wise relative variograms and their models estimated for a Al2O3, b SiO2 and c Fe grades using the
blast-hole sample assays

Applied Earth Science (Trans. Inst. Min. Metall. B) 2010 VOL 119 NO 3 141
Abzalov et al. Grade control at Yandi mine, Pilbara region, Australia

12 Experimental pair-wise relative variograms and their models estimated for a Al2O3, b SiO2 and c Fe grades using the
RC and diamond hole samples composited to 10 m benches.

Table 4 Models of the pair-wise relative variograms of Al2O3, SiO2 and Fe grades estimated using the blast-hole samples

Modelled variable Nested structure Sill Range Comments

Al2O3 Major axis (Azi 050) Minor axis (Azi 140)


Nugget 0.045 Nugget effect
Spherical 0.0435 20 12 Local component
Exponential 0.028 140 140
Linear 0.01 1400 55 Regional drift

SiO2 Major axis (Azi 050) Minor axis (Azi 140)


Nugget 0.011 Nugget effect
Spherical 0.0215 25 25 Local component
Exponential 0.0135 175 140
Linear 0.0027 1400 57 Regional drift

Fe Major axis (Azi 050) Minor axis (Azi 140)


Nugget 0.00034 Nugget effect
Spherical 0.00065 35 18 Local component
Linear 0.0001 800 60 Regional drift

Table 5 Models of the pair-wise relative variograms of Al2O3, SiO2 and Fe grades estimated using the RC and diamond
hole samples composited to 10 m benches

Modelled variable Nested structure Sill Range Comments

Al2O3 Major axis (Azi 000) Minor axis (Azi 090)


Nugget 0.065 Nugget effect
Spherical 0.05 25 50 Local component
Exponential 0.035 175 160
Linear 0.01 1400 160 Regional drift

SiO2 Major axis (Azi 000) Minor axis (Azi 090)


Nugget 0.011 Nugget effect
Spherical 0.0215 25 50 Local component
Exponential 0.059 175 160
Linear 0.0048 1400 140 Regional drift

Fe Major axis (Azi 000) Minor axis (Azi 090)


Nugget 0.0003 Nugget effect
Spherical 0.00075 120 120 Local component
Linear 0.0001 1500 160 Regional drift

142 Applied Earth Science (Trans. Inst. Min. Metall. B) 2010 VOL 119 NO 3

You might also like