The Importance of Sampling in Grade Control

R J Holmes1

ABSTRACT
A wide variety of process streams are sampled for grade control in the
mining industry. However, in many instances the overall precision of the
resultant analyses is inadequate for the intended purpose, ie grade control
and optimisation of resource utilisation. Consequently, special attention
needs to be given to the sampling, sample preparation and analysis
procedures in place to ensure that meaningful grade control decisions can
be made based on the analyses generated. For example, it is virtually
impossible to control a process to within, say, ±0.2 per cent if the overall
precision of the analyses is no better than 0.5 per cent. Hence, precision
targets appropriate for the task need to be established and the overall
precision achieved in practice monitored and controlled on an ongoing
basis.

INTRODUCTION AND BACKGROUND

Samples are taken from a range of locations in the mineral
industry, including percussion and diamond drill holes for
mineral exploration and resource evaluation, as well as blast
holes, feed and product streams, conveyors, wagons and FIG 1 - Sampling from the tops of railway wagons ignores material
stockpiles for grade control and commerical purposes. The focus at the bottom of the wagon and hence does not provide a
of this paper is on sampling in process plants for grade control, representative sample.
for which there are two important requirements:
• the samples need to be free of significant bias, and
• the overall precision of the resultant analyses needs to be
appropriate for the grade control task at hand.
Unfortunately, not all sampling systems and procedures that
are used in the mineral industry for grade control satisfy the
above two criteria. Quite often they do not conform to the
fundamental rule for correct sampling and sample preparation
that ‘all parts of the material being sampled must have an equal
probability of being collected and becoming part of the final
sample for analysis’ (Gy, 1982; Pitard, 1993, 2005; Holmes,
2004, 2005, 2007). For example, sampling from the tops of
conveyor belts and railway wagons, sampling from the sides of
stockpiles or taking samples from only part of an ore stream are
unacceptable for the following reasons:
• For conveyor belts, both vertical and horizontal segregation
of ore occurs due to the action of the idlers and the manner in
which the ore is fed onto the conveyor, so samples taken
from the top of the conveyor or only one side of the ore
stream will be biased. The same considerations apply to
sampling from the tops of railway wagons (see Figure 1).
• For stockpiles, the coarser particles tend to rill down the side
of the stockpile, so samples taken from the side of the
stockpile will be biased towards the coarser particles and
cannot possibly provide any information on the ore in the
main body of the stockpile (see Figure 2). FIG 2 - Sampling from the side of a stockpile does not provide
Regarding the overall precision of grade measurements, it is information on the grade of material in the main body of the
surprising in this day and age with the plethora of workshops and stockpile and hence does not provide a representative sample.
training courses on quality assurance and quality control
(QA-QC) how little thought is often given to the precision collected during shiploading (on which grade control decisions
required and how seldom it is actually measured and known to are made) is no better than about 0.3 per cent Fe, ie ±0.6 per cent
grade controllers and quality control personnel. For example, it is Fe at the 95 per cent confidence level. Any changes that are
virtually impossible to control the grade of an iron ore shipment made to the loading plan may merely be a reaction to
to within, say, ±0.25 per cent Fe if the overall precision (one ‘measurement noise’ rather than real changes in grade. This is
standard deviation, σ) of the analyses of individual subsamples illustrated in Figure 3 for a target grade of 62 per cent Fe and
shows the distribution of grade measurements for overall
1. FAusIMM, Theme Leader – Iron Ore, CSIRO Minerals, Technology precisions (1σ) of 0.1 and 0.3 per cent Fe. There is some chance
Court, Pullenvale Qld 4069. Email: [email protected] of controlling the grade to 62 ± 0.25 per cent Fe when the overall

Sampling Conference Perth, WA, 27 - 29 May 2008 77

R J HOLMES

precision of grade measurements is 0.1 per cent Fe (solid line),

but the task is virtually impossible if the overall precision is
considerably poorer at 0.3 per cent Fe (dashed line). In the latter
case, a measured grade of 61.5 per cent Fe may be due to an
actual change in grade (requiring action) or simply due to
measurement variance (no action required). The most reliable
solution to this problem is to ensure that the measurement
standard deviation is considerably smaller than the grade control
limits being targeted.

StDev
0.1
0.3

FIG 4 - Correctly designed cross-stream primary cutter for

sampling a high capacity iron ore stream.

• The sample cutter must be non-restrictive and self-clearing,

discharging completely each increment without any reflux,
overflow or hang-up.
• The geometry of the cutter opening must ensure that the
61.0 61.5 62.0 62.5 63.0
Target Grade (%Fe) cutting time at each point in the stream is equal, ie
linear-path cutters must have parallel cutter lips, while Vezin
cutters must have radial cutter lips.
FIG 3 - Distribution of iron ore grade measurements corresponding
to overall analysis precisions (1σ) of 0.1 per cent Fe (solid line) • The cutter should intersect the stream either in a plane normal
and 0.3 per cent Fe (dashed line). to, or along an arc normal to, the mean trajectory of the
stream.

BIAS MINIMISATION AND CUTTER DESIGN • The plane of the cutter lips must not be vertical or near
vertical.
The bad news is that once bias has been introduced into a • The cutter must travel through the stream at a uniform speed,
measurement chain, no amount of replicate sampling and accelerating up to its cutting speed before entering the stream
analysis will eliminate it. The end result of replicate sampling and then decelerating to a stop after leaving the stream.
and analysis is only a more precise estimate of the incorrect
grade, so it is critical that bias be at least minimised or preferably • The cutter aperture must be at least three times the nominal
eliminated at the outset. Sources of bias that can be eliminated top size (d) of the material being sampled to prevent
through correct design of sampling and sample preparation preferential loss of the larger particles, subject to a minimum
systems and procedures include sample spillage, sample of 10 mm for slurries and fine dry solids, and a minimum of
50 mm for wet solids to minimise ‘bridging’ of the cutter
contamination, incorrect delimitation and extraction of
aperture.
increments and preferential exclusion of specific size fractions,
eg loss of the coarser or finer particles, while sources of bias that • The cutter speed should not exceed 0.6 m/s at the minimum
can be minimised but not completely eliminated include change cutter aperture of 3d to avoid extraction bias (Gy and Marin,
in moisture content, size degradation and dust loss. Of these 1978). However, if the cutter aperture is increased above this
factors, incorrect delimitation and extraction of increments are minimum, the maximum cutter speed can be increased above
very common causes of bias, but these can be addressed by this minimum subject to an absolute maximum of 1.2 m/s.
correct design of the sample cutter. • Bucket-type cutters must have sufficient capacity to
In a processing plant, the best method of collecting unbiased accommodate the increment mass obtained at the maximum
sample ‘increments’ is to sample a moving stream at a transfer flow rate of the stream without any reflux or overflow from
point between two conveyor belts or at the end of a slurry pipe the cutter aperture.
where the full stream can be intercepted at regular intervals and • No materials other than the sample must be introduced into
increments collected by taking a full cross-section of the stream the cutter or the sample delivery chute.
with a sample cutter. This approach should also be applied where
the quality of ore in a railway wagon or a stockpile is required,
• If a belt scraper is required to remove material adhering to
the belt, the scraped material must fall within the area
by sampling the ore from a moving stream when the wagon is
traversed by the cutter.
being loaded or unloaded or when the stockpile is being built-up
or reclaimed. Sample cutters must satisfy the following design These design criteria are also specified in the key ISO
rules to eliminate increment delimitation and extraction errors standards for sampling mineral commodities, eg ISO 3082
(Holmes, 2004, 2005, 2007): (Sampling of iron ores), ISO 12743 (Sampling of copper, lead,
zinc and nickel concentrates) and ISO 13909 (Sampling of coal
• The cutter must collect a complete cross-section of the and coke).
stream, both the leading and trailing edges of the cutter When sampling slurries, the cutter lips must be sharp and the
completely clearing the stream at the end of each traverse sample cutter streamlined to minimise turbulence as it cuts the
(see Figure 4). Cutters that collect only part of the stream on slurry stream (Bartlett, 2002). In addition, loss of sample due to
a continuous or periodic basis, eg ‘shark fin’ cutters, dribbling from the end of pipes must be avoided, so the
‘pressure pipe’ samplers, T-pieces or ‘bleeds’ from the side of dimensions of the cutter need to be generous enough to intersect
a pipe for sampling slurries, are not acceptable. all material exiting the pipe (see Figure 5).

78 Perth, WA, 27 - 29 May 2008 Sampling Conference

 THE IMPORTANCE OF SAMPLING IN GRADE CONTROL

• ascertain the ‘quality variation’ of the material being sampled

for determining the number of primary increments needed to
obtain the required sampling precision;
• determine the sampling interval in tonnes for mass-basis
sampling and in minutes for time-basis sampling; and
• establish the procedure for combining increments into sublot
samples or a gross sample to achieve the required overall
precision.
The above procedure may require several iterations and
balancing of the separate contributions from the sampling,
sample preparation and analysis variances to achieve the required
overall precision. The method of taking account of these variance
contributions is slightly different in the various ISO standards for
mineral commodities, but, apart from the method for determining
‘quality variation’, eg the variance of primary increments for
coal and copper, lead, zinc and nickel concentrates (ISO 13909
and ISO 12743) and the ‘within strata’ quality variance for iron
ore (ISO 3084), the differences relate largely to terminology and
definitions of variance components. In the case of iron ore for
example (ISO 3082), the overall variance, σ 2SPM , is given by:

σ 2SPM = σ 2S + σ 2P + σ 2M (1)

where:
σ 2S = sampling variance
FIG 5 - Cross-stream primary cutter for sampling slurries.
σ 2
P = sample preparation variance
While both cross-stream and cross-belt cutters are used in σ 2M = measurement variance, including selection of the final test
industry for sampling moving streams, particularly in the coal portion
industry, cross-stream cutters are strongly recommended because Hence, assuming the quality variation, σ W, and the precisions
cross-belt cutters tend to leave a layer of material on the of sample preparation and analysis are known, the overall
conveyor belt if the contour of the conveyor belt does not match precision (1σ) can be calculated using the following equation:
the trajectory of the cutter and/or the skirts at the bottom of the
cutter are not correctly adjusted as they wear out (Docherty,
2005). In addition, maintenance personnel sometimes increase σ 2P σ 2M σ 2W σ 2P σ 2M
the gap between the cutter and the conveyor to eliminate possible σ SPM = σ 2S + + = + + (2)
k kr n k kr
damage to expensive conveyor belts. For these reasons,
increment extraction is incorrect and hence cross-belt cutters can
be seriously biased, because the material on the bottom of the where:
belt is usually different in grade from the bulk of the material on n = number of primary increments taken from the ‘lot’
the conveyor belt. Another problem with cross-belt cutters is that
it is virtually impossible to check them visually to determine k = number of sublot samples prepared for analysis
whether they are performing correctly, and visual inspections are r = number of replicate analyses of the analysis samples
far more effective for checking performance than conducting bias
tests due to the sheer cost of conducting statistically sound bias As an example of the application of Equation 2, consider a
tests. 100 000 t stockpile being sampled while it is being built up in
accordance with ISO 3082, which specifies 50 primary
increments (n = 50) for small quality variation. If the quality
ESTABLISHING AN APPROPRIATE variation of the ore being stockpiled is, say, 0.6 per cent Fe
SAMPLING REGIME (small quality variation), the stockpile is sampled as ten sublots
The principal steps in establishing an appropriate sampling of 10 000 t each (k = 10), the precisions of sample preparation
regime are as follows: and analysis are each 0.1 per cent Fe, and the ten sublot samples
are separately constituted, prepared and analysed in duplicate
• identify what needs to be sampled, ie the ‘lot’, which could (r = 2), then σ SPM = 0.09 per cent Fe. This is quite acceptable for
be a day’s production, a stockpile build, a shipment or the grade of the entire stockpile of 100 000 t and well within the
10 000 t of ore, concentrate or coal; precision targets specified in ISO 3082.
• specify the purpose of sampling, eg determination of the On the other hand, Equation 2 indicates that the overall
average grade of a day’s production or the grade of precision of the analyses of individual sublot samples (n = 5,
individual increments or sublots for making grade control k = 1 and r = 2) is σ SPM = 0.29 per cent Fe, so difficulties will be
decisions; experienced if the objective is to use these analyses to control the
• specify the required sampling precision and overall precision final grade of the stockpile to within, say, ±0.25 per cent Fe as
of sampling, sample preparation and analysis, taking into already illustrated in Figure 3. The best solution to this problem
account the purpose of sampling; is to take more primary increments for constituting each sublot
• ascertain the nominal top size of the material to be sampled sample. The alternative is to improve the precisions of sample
and hence the dimensions of the sample cutter and mass of preparation and analysis, although in the above example this will
increment; result in only a small improvement in the overall precision.

Sampling Conference Perth, WA, 27 - 29 May 2008 79

R J HOLMES

SYSTEM AUDITING AND PRECISION CHECKS Finally, sampling installations need to be equipped with
duplicate sampling facilities for determination and ongoing
Given the importance of good sampling, sample preparation and monitoring of the precision achieved in practice. For this
analysis systems for grade control, ongoing verification that purpose, the sampling installation needs to be able to direct
sampling systems are performing correctly is critical (François- alternate primary increments to duplicate samples A and B, each
Bongarçon and Gy, 2002) and reasonably comprehensive check of which is then prepared in duplicate and each final analysis
lists are now being incorporated into ISO standards, eg the latest sample analysed in duplicate, resulting in up to eight analyses for
version of ISO 3082 which should be published in 2008. Key each lot or sublot depending upon the sample division scheme
items that need to be incorporated in such system checks and
used. This process is then repeated for at least 20 lots or sublots,
audits include:
thereby enabling the sampling, sample preparation, analysis and
• size and geometry of cutter apertures, including checking overall precisions to be determined, eg using the procedures
that cutters take a complete cross-section of the stream, specified in ISO 3085, ISO 13292 and ISO 13909: Part 7. Unless
• worn and/or missing cutter lips, these separate precisions are measured, it is impossible to
intelligently tailor a sampling regime to meet grade control
• build-up and/or blockage of cutter apertures, requirements as discussed above. It is also poor practice to
• reflux from the cutter aperture, blindly submit samples for analysis and take action on the results
• cutter speed and uniformity while cutting the stream, without knowing their precision and reliability.
An example of precision measurements for the Fe content of a
• number of cuts,
large number of Australian iron ore shipments ranging up to
• ingress of extraneous material when the cutter is in the 190 000 t is provided in Table 1 for two different ports and five
parked position, different ore types. Not only are the overall precision values
• holes in chutes and bins resulting in sample loss, excellent and far better than the minimum requirements set down
in ISO 3082, but it is also clear that the sampling characteristics
• particle size, and of ore type five are significantly better than for ore types one to
• increment/sample mass at each stage of sampling and sample four and in all cases there is a reasonable balance between the
preparation. sampling, sample preparation and measurement precisions. The
An example of a primary cutter with a blocked cutter aperture raw data used for determining these precision values were
that was discovered during a sampling audit is shown in Figure 6, generated via routine duplicate sampling of iron ore shipments,
which clearly was a major problem and needed to be rectified thereby providing an ongoing check of the precision actually
immediately. achieved in practice at minimal additional cost.

TABLE 1
Precision achieved in practice for the Fe content of five different
iron ore types at two different Australian ports for shipments
varying from about 20 000 to 190 000 t.
Ore type Port σS σP σM σ SPM
(%Fe) (%Fe) (%Fe) (%Fe)
1 A 0.051 0.042 0.028 0.071
B 0.080 0.035 0.031 0.093
2 A 0.053 0.034 0.024 0.067
B 0.082 0.043 0.028 0.097
3 A 0.055 0.034 0.027 0.070
B 0.068 0.038 0.027 0.082
4 A 0.059 0.043 0.027 0.077
B 0.068 0.037 0.021 0.080
5 A 0.023 0.024 0.023 0.040
FIG 6 - Primary cutter with blocked cutter aperture. B 0.021 0.027 0.022 0.041

A useful tool for checking the design and operation of sample CONCLUSION
cutters is the ‘extraction ratio’ (Docherty, 2005). This is the ratio
of the actual increment mass collected to the theoretical Special attention needs to be given to sampling, sample
increment mass, and if this ratio is significantly less than one preparation and analysis procedures in mining and mineral
then the cause needs to be identified and corrective action taken. processing operations to ensure that the grade information
The extraction ratio should be determined as a function of flow generated is unbiased and sufficiently precise for meaningful
rate, because problems with reflux for example become more grade control decisions to be made. The basic rule for correct
serious as the flow rate increases. sampling and sample preparation is that all parts of the material
Bias testing should also be conducted when a new sampling being sampled must have an equal probability of being collected
system has been installed or major modifications have been made and becoming part of the final sample for analysis. If this is not
to an existing system, eg using the procedures in ISO 3086, ISO the case, then bias is easily introduced that cannot be eliminated
12744 and ISO 13909: Part 8. However, bias tests are very by simply averaging replicate measurements. Consequently, it is
expensive to conduct and comprehensive audits of sampling critical that sampling equipment is carefully designed in
installations to confirm that their design and operation conforms accordance with correct sampling principles to minimise and
to correct sampling principles are in many instances a more preferably eliminate bias. In addition, the required overall
effective alternative. precision of grade measurements must be established taking into

80 Perth, WA, 27 - 29 May 2008 Sampling Conference

 THE IMPORTANCE OF SAMPLING IN GRADE CONTROL

account their intended use in grade control, and then sampling International Organization for Standardization (ISO), 1998. ISO 3084,
regimes designed accordingly to satisfy these precision Iron ores – Experimental methods for evaluation of quality variation
(International Organization for Standardization: Geneva).
requirements. To achieve this, sampling facilities need to be
designed with duplicate sampling capabilities to enable the International Organization for Standardization (ISO), 2000. ISO 3082,
Iron ores – Sampling and sample preparation procedures
precisions of sampling, sample preparation and analysis, as well as (International Organization for Standardization: Geneva).
overall precision, to be determined and monitored on an ongoing International Organization for Standardization (ISO), 2001. ISO 13909,
basis. Parts 1-8. Hard coal and coke – Mechanical sampling (International
Organization for Standardization: Geneva).
REFERENCES International Organization for Standardization (ISO), 2002. ISO 3085,
Iron ores – Experimental methods for checking the precision of
Bartlett, H E, 2002. Design of primary samplers for slurries in sampling, sample preparation and measurement (International
concentrators and statistical methods for measuring components of Organization for Standardization: Geneva).
variance in sampling, J South African Institute of Mining and International Organization for Standardization (ISO), 2006a. ISO 3086,
Metallurgy, 102(8):485-490. Iron ores – Experimental methods for checking the bias of sampling,
Docherty, J, 2005. Mechanical sample plants, in Proceedings Second sample preparation and measurement (International Organization for
World Conference on Sampling and Blending, pp 83-93 (The Standardization: Geneva).
Australasian Institute of Mining and Metallurgy: Melbourne). International Organization for Standardization (ISO), 2006b. ISO 12743,
François-Bongarçon, D F and Gy, P, 2002. Critical aspects of sampling in Copper, lead, zinc and nickel sulphide concentrates – Sampling
mills and plants: a guide to understanding sampling audits, J South procedures for determination of metal and moisture content
African Institute of Mining and Metallurgy, 102(8):481-484. (International Organization for Standardization: Geneva).
Gy, P M, 1982. Sampling of Particulate Materials – Theory and Practice, International Organization for Standardization (ISO), 2006c. ISO 12744,
second edition (Elsevier: Amsterdam). Copper, lead, zinc and nickel concentrates – Experimental methods
Gy, P M and Marin, L, 1978. Unbiased sampling from a falling stream of for checking the bias of sampling (International Organization for
particulate material, Int J Mineral Processing, 5(1978):297-315. Standardization: Geneva).
Holmes, R J, 2004. Correct sampling and measurement – The foundation International Organization for Standardization (ISO), 2006d. ISO 13292,
of metallurgical accounting, Chemometrics and Intelligent Copper, lead, zinc and nickel concentrates – Experimental methods
Laboratory Systems, 74:71-83. for checking the precision of sampling (International Organization
Holmes, R J, 2005. Design of sample plants – Getting it right first time, for Standardization: Geneva).
in Proceedings Second World Conference on Sampling and Blending, Pitard, F F, 1993. Pierre Gy’s Sampling Theory and Sampling Practice,
pp 103-110 (The Australasian Institute of Mining and Metallurgy: second edition (CRC Press, Inc: Boca Raton).
Melbourne). Pitard, F F, 2005. Sampling correctness – A comprehensive guideline, in
Holmes, R J, 2007. Best practice in sampling iron ore, in Proceedings Proceedings Second World Conference on Sampling and Blending,
Third World Conference on Sampling and Blending, Porto Alegre, pp 55-66 (The Australasian Institute of Mining and Metallurgy:
Brazil, 23-25 October, pp 416-429. Melbourne).

Sampling Conference Perth, WA, 27 - 29 May 2008 81