Maintaining Sample Data Quality

Through Robust Quality Assurance and

Quality Control Protocols
J Graindorge1

ABSTRACT
Successful evaluation and exploitation of resources and reserves relies on the collection of high
quality sample data. The quality of the sample data must be maintained through the application
of a robust quality assurance and quality control (QAQC) program, including documentation
of sampling procedures, training of sampling personnel and the regular monitoring of control
samples. However, the application of QAQC programs in the minerals industry is often poor and
the purpose misunderstood. It is all too common to have good quality control data collected, but no
analysis work carried out until after drilling is completed. In a lengthy drill program, this can lead
to problems not being discovered and corrected as they happen. Moreover, rectifying problems
afterwards will be costly in terms of time and money, if not impossible.
The consequences of poor quality sample data are many and include such effects as over- or
underestimation of block grades, misclassification of ore and waste, and lower confidence in
the resource/reserve models. The purpose of this paper is to highlight, using case studies, the
importance of good QAQC practices to monitor the quality of sampling and laboratory sample
preparation and analysis.

INTRODUCTION
To ensure that reliable sampling and assaying data is obtained, adequate QAQC program, can result in effects such as bias
a robust quality assurance and quality control (QAQC) program in the resource grade estimates, misclassification of ore and
is required and should be an important part of any sampling waste during grade control, poor reconciliation (Noppé,
and assaying program. There are many stages a sample must 2004), elevated nugget effect (Voortman, 1998) and an overall
go through in order to obtain an assay result. Errors and faults lack of confidence in the resource and reserve models. In the
can occur at any of these stages, so it is important to have author’s experience, QAQC is often perceived as something
systems in place that minimise these errors and faults, and that must be done to satisfy reporting requirements, such as
action procedures for when errors are identified. the JORC Code. However, the primary aim of a QAQC system
The theory of sampling (Pitard, 1993 and Gy, 1998) provides is to ensure that sample data is of consistently high quality.
an insight into the causes and magnitude of errors and This is achieved by providing a means of monitoring the
variability that may occur during sampling and assaying. sample data quality and resolving errors in a timely fashion
(ie as they occur).
Unfortunately, these errors can never be zero due to the
heterogeneity of the material and the limitations of sampling The general principles of QAQC are covered in several
and analytical devices. The sampling errors, however, can be studies (eg Greig and Cook, 1998; Leppard, 1998; Voortman,
minimised and controlled through correct sampling practices. 1998; Noppé, 2004; Simon and Gossan, 2008). These studies
tend to discuss the importance of high quality data, however
The sample quality must be monitored and maintained
very few case studies exist which highlight the consequences
throughout the sampling and sample preparation chain. The
of poor QAQC practices, likely due to the negative nature of
current JORC Code (JORC, 2004) requires companies to
these types of case studies.
disclose ‘… any relevant information concerning a mineral
deposit that could materially influence the economic value One such case study is the Gosowong gold mine in Indonesia,
of that deposit …’. Given that samples form the basis of any where Carswell et al (2009) describe the use of duplicate
resource or reserve evaluation, this includes information sampling during grade control to improve reconciliation
regarding sample quality and reliability, which should be results. Analysis of duplicate samples from the run-of-mine
represented by appropriate resource classification. The (ROM) stockpiles showed an unacceptably poor precision,
establishment of robust QAQC procedures and checks provides especially for very high-grade samples. The poor precision
a means of monitoring the reliability of the sample results. of the duplicates initiated a review of the ROM sampling
The assessment of mineral deposits relies on high quality procedure and subsequently a number of improvements were
sample data to make potentially costly decisions. Poor quality made, including increasing the sample size and ensuring
sample data, typically caused by a lack of training and an that larger particles were not under-sampled. Further QAQC

1. MAusIMM(CP), Senior Consultant, Snowden Mining Industry Consultants, 87 Colin Street, West Perth WA 6005. Email: [email protected]

SAMPLING CONFERENCE / PERTH, WA, 11 - 12 MAY 2010 111

 J GRAINDORGE

procedures were also implemented, the result of which was a the same lot. The precision of sampling is usually assessed
reduction in the end of month reconciliation bias, as measured using duplicate samples.
by the ROM sampling compared to the mill production, from The term accuracy, in a sampling and assaying context,
7.4 per cent down to 3.4 per cent (Carswell et al, 2009). describes how close the average of repeat samples is to the
Voortman (1998) outlines a case where gold grades from ‘true’ value of the lot. However, the ‘true’ value is a theoretical
a volcanogenic massive sulfide deposit were believed to concept and can never be actually measured definitively (Simon
be biased, due to analytical errors caused by calibration and Gossan, 2008) due to the presence of heterogeneity and
problems. The sample pulps were subsequently re-assayed analytical errors. Reference materials, where the ‘true’ value
by the original laboratory and were also submitted to two is effectively known through repeated assaying, are used to
other laboratories. Analysis of the re-assay results from assess analytical accuracy. When inaccuracies occur, samples
the three laboratories showed that the gold grades were will often be biased, either due to the sampling system/
underestimated by the original laboratory by approximately process, or due to analytical errors.
7.5 per cent (Voortman, 1998). The implication of this study is
that, as the geostatistical grade estimates are fundamentally Duplicate samples
based on the underlying sample data, any resource estimate Assessing the precision of assay data is an important part of
(and subsequent reserve calculation) based on the biased QAQC and is primarily determined by collecting and analysing
sample data would also underestimate the gold content by duplicate samples. A duplicate is a second sample split from
around 7.5 per cent. the original sample. Duplicate samples are prepared and
analysed as separate samples and should be issued with
QUALITY ASSURANCE/QUALITY CONTROL a unique sample number. Given the purpose of duplicate
PRINCIPLES samples, the duplicate must be correctly split from the original
sample and must be prepared and analysed in the same way
There are two components to a QAQC system – quality (Simon and Gosson, 2008).
assurance and quality control. Quality assurance (QA) refers
Statistical analysis of duplicate data utilises a variety of tools,
to the protocols and procedures that ensure that sampling and
which compare the duplicate sample to the original sample,
assaying is completed to the required quality (Noppé, 2004).
either directly or by calculating the difference between the
Quality control (QC), however, is the use of control samples
duplicate pairs.
and statistical analysis to ensure that the assay results are
reliable (Noppé, 2004). In order for QC to be effective, the Ideally, duplicate samples should be collected at each size
analysis must be completed in a timely manner, ideally as reduction stage of the sample preparation chain (Figure 1).
soon as assay results are received. Sampling errors are encountered at all stages in the sample
preparation chain and are cumulative. Consequently, the
The aim of sampling and assaying is to estimate the grade of variability between field duplicates is expected to be higher
a large volume of material based on a small volume of material than the variability between other duplicate sample types as
collected from that lot. The reality of sampling, however, is field duplicates are collected the earliest in the chain.
that the character of a sample can never be exactly the same
as that of the lot due to a variety of sampling errors caused by Reference materials
both the nature of the material being sampled and the design Reference materials are samples, usually a pulp, with a
and implementation of the sampling system (Pitard, 1993 and known grade and variability. Analytical accuracy and bias
Gy, 1998). This difficulty in taking a sample, which is fully can therefore be assessed by comparing the assay results of
representative of the lot, results in variations in assay results the reference material to the expected grade. The results are
for repeat samples taken from the lot, and can be compounded typically monitored using control charts, which plot the assay
during the subsampling and assaying process due to sampling results of the reference material over time. This not only allows
errors at each stage. for an assessment of the deviation from the expected grade,
but means that trends, such as from analytical drift, can also
The variation in repeat or duplicate samples taken from be identified and corrected in a timely manner. Control limits,
a lot can be quantified in terms of the level of precision and usually defined by a standard deviation, are set on the chart to
accuracy attached to the sample assays. define the region in which the assay results are considered to be
Simon and Gossan (2008) describe precision as the closeness in statistical control.
or agreement of assay results obtained under stipulated Analytical bias in reference material results can also be
conditions. In other words, the spread or repeatability of assessed using control charts. If no bias exists then the average
sampling results when the sampling process is repeated from of the assay results should be very close to the expected value.


FIG 1 - Sample preparation chain showing duplicate sample points.

112 SAMPLING CONFERENCE / PERTH, WA, 11 - 12 MAY 2010

MAINTAINING SAMPLE DATA QUALITY THROUGH ROBUST QUALITY ASSURANCE AND QUALITY CONTROL PROTOCOLS

Similarly, if no bias exists, the error or difference between the Case study 1 – Reference materials
results and the expected value should average out to be close
During exploration and resource definition of a haematite iron
to zero. In other words, the results should fluctuate randomly
ore deposit in Australia, the company was using commercial
above and below the expected value, with the positive and
reference materials as part of its QAQC protocols to monitor
negative differences cancelling each other out.
the analytical accuracy of the laboratory. Three reference
The selection of appropriate reference materials for a project materials were utilised during the 2007 to 2008 reverse
must be based on the geology and chemical characteristics of circulation (RC) drilling program. The reference materials
the deposit under evaluation. Reference materials that are
used were 603-1, 606-1 and DC14007a, with Fe grades of
significantly different in composition from routine samples,
53.65 per cent, 59.66 per cent and 61.73 per cent respectively.
may affect the reliability and validity of the results. Similarly,
Routine samples were analysed for Fe by X-ray fluorescence
the assaying methodology employed should be the same as
(XRF) at a commercial laboratory in Australia.
used to certify the reference material. For example, if the
expected gold grade of a reference material is based on fire The company QAQC protocol for the 2007 to 2008 drilling
assay results, it follows that this particular reference material program included the use of field duplicates along with the
can only be used to assess the accuracy of fire assay results. inclusion of samples of reference material in the assay batches.
Using aqua regia, for instance, would result in an apparent The QAQC data was reviewed prior to resource estimation to
analytical bias (aqua regia typically does not achieve full gold ascertain the precision and accuracy of the RC sample data.
extraction); however the bias is due to the difference in the Analysis of the field duplicates showed that the precision
analytical techniques, rather than laboratory errors. of the field sampling was within acceptable limits (majority
Whether commercially available or in-house generated of duplicate pairs have a relative difference of less than five
reference materials are used, the reference materials need per cent). The reference material data, however, showed
to be selected to be representative of the grades and material significant analytical bias in the Fe grades.
types normally encountered within a deposit. Typically, at The results for the three reference materials are summarised
least three to four reference materials should be used of in Table 1. Most of the results for reference materials 603-1
varying grades (low-grade, high-grade and one or two mid- and DC14007a fall within acceptable limits (±3 standard
grade). deviations from the expected grade), however there is a bias of
0.10 to 0.13 per cent Fe present, with the results consistently
CASE STUDIES reporting below the expected Fe grade of the reference
The following case studies are used to illustrate the importance material. For reference material 606-1, the bias is opposite,
of good QAQC practices and the timely analysis of the data with the results averaging 0.43 per cent Fe above the expected
collected. The nature of the case studies necessitates that the grade (Figure 2). This bias in the XRF results was deemed
companies and laboratories involved remain anonymous. unacceptable and further investigation was warranted.

TABLE 1
Iron ore reference material results from the 2007 to 2008 RC drilling campaign.

Reference material Results

ID Expected value % Fe Standard deviation % Fe Number Mean % Fe Diff % Fe
606-1 59.66 0.14 39 60.09 0.43
603-1 53.65 0.15 38 53.55 -0.10
DC14007a 61.73 0.10 37 61.60 -0.13

FIG 2 - Control chart for standard 606-1.

SAMPLING CONFERENCE / PERTH, WA, 11 - 12 MAY 2010 113

 J GRAINDORGE

Initial observations provided mixed messages. For the poor precision, along with a bias, with the duplicate samples
low-grade and high-grade reference materials, the analytical reporting higher grades than the original samples.
bias is negative (ie assay results below the expected value of Figure 3 shows a quantile-quantile (Q-Q) plot comparing
the reference material), however the mid-grade reference the original sample grades to the field duplicate sample
material showed a positive bias in the XRF results. A bias
grades and illustrates the bias present. For samples below
caused by analytical errors (eg poor calibration) within the
approximately 1 g/t Au, the field duplicates report a grade that
laboratory would be expected to be consistently over- or
is, on average, approximately 10 – 50 per cent higher than the
under-reporting the grade. Hence, the biases exhibited in the
reference material results are unlikely to be caused entirely by original sample.
a laboratory error.
Rather than initiate a potentially expensive round of re-
assaying, the reference materials were reviewed to assess the
validity of the expected grades, along with the properties of
the source material.
The three reference materials were found to be composed of
material sourced from haematite iron ore, although the exact
source is unknown. Reference materials 603-1 and 606-1 were
certified by Institut de Recherches de la Sidérurgie (IRSID)
in France in the late 1960s using volumetric techniques.
Reference material DC14007a was certified by the National
Analysis Centre for Iron and Steel in China using various
volumetric techniques.
The analytical technique used by the company for routine
analysis of iron ore samples, including the reference materials,
was XRF and it is likely that the biases observed are partly due
to the difference in analytical techniques. Moreover, at least
two of the reference materials were certified in the late 1960s
(approximately 45 years ago) and hence are likely to have
absorbed moisture, changing the composition slightly.
In this case, rather than analytical errors in the laboratory,
the reference materials utilised by the company were not
suitable for the purpose of assessing analytical accuracy of the
routine drillhole samples. Moreover, very little analysis of the
FIG 3 - Q-Q plot comparing gold grades of the original samples with the field
QAQC results was completed during the drilling campaigns
duplicates.
and consequently the reliability of the drillhole data set from
2007 to 2008 is brought into question. Had adequate analysis Measurements of the sample weights, both original and
been conducted, the biases observed in the reference material field duplicate, were routinely collected during drilling and
assays would have been identified early on and the confidence the drillhole database included information detailing which
in the assay data restored. splitter was used during sampling (cone or four-tiered riffle
The company has since changed to utilising more recent splitter). A total of 242 duplicate samples were collected using
reference materials to evaluate the analytical accuracy, which the cone splitter, while 128 duplicate samples were collected
are certified using XRF. using the four-tiered riffle splitter.
The field duplicate samples collected using the cone splitter
Case study 2 – Field duplicate sample show similar weights to the original samples, with no bias
collection evident. However, the four-tiered riffle splitter shows a strong
constant bias, with the duplicate samples being significantly
During 2006 and 2007, a gold deposit in central Western
lower in weight compared to the original sample (Figure 4).
Australia was sampled using RC drilling. Gold mineralisation
The field duplicate samples average approximately 0.86 kg
is associated with disseminated pyrite and quartz-pyrite
lighter than the original sample.
veining within a package of altered mafic schist, chert and
banded iron formation. Aeromagnetic interpretations suggest The difference in sample weights was primarily due to the
that the mineralisation is associated with the hinge zone of design and use of the four-tiered riffle splitter employed
an antiformal structure. The mineralisation is low grade in during the 2006 to 2007 sampling program (Figure 5a and
nature, with gold grades averaging approximately 1 g/t Au. Figure 5b). The splitter, along with the cyclone, is rigidly
attached to a trailer, with no mechanism to level the splitter.
At the time of drilling, the company’s QAQC procedures
The drill cuttings, which were predominantly dry, were collected
included the use of field duplicate samples, collected every
from the bottom of the cyclone in large plastic sample bags. The
20 samples, along with the use of commercial gold reference
materials, which were included in the sample batches at a rate material was then passed through the four-tiered riffle splitter,
of 1:20. Depending on the drill contractor being employed, with the sample poured directly from the plastic bag.
samples were split from the RC drill cuttings using either a A riffle splitter, when used correctly (Pitard, 1993), is
cone splitter or a four-tiered riffle splitter. designed to split a particulate material in half (ie 50:50).
Prior to resource estimation, the QAQC data was reviewed Ideally, riffle splitters should be single tiers with the material
to assess the precision and accuracy of the sample and assay passed through the splitter multiple times to achieve the
data. Analysis of the reference material results showed desired split. For practical reasons, riffle splitters are often
acceptable analytical accuracy was being achieved for gold tiered to reduce the splitting time and to avoid safety issues
assays. However, an analysis of the field duplicate data showed arising from manual handling. However, a tiered riffle

114 SAMPLING CONFERENCE / PERTH, WA, 11 - 12 MAY 2010

MAINTAINING SAMPLE DATA QUALITY THROUGH ROBUST QUALITY ASSURANCE AND QUALITY CONTROL PROTOCOLS

Case study 3 – Laboratory procedures

During exploration of a haematite iron ore project in central
Africa in 2009, a company was using reference materials
to assess the analytical accuracy of the laboratory assaying.
An independent commercial laboratory in Africa was used
for routine XRF analysis of the exploration samples. The
reference materials used were GIOP-17, GIOP-18 and GIOP-
19, produced and certified by Geostats Pty Ltd in Western
Australia. These pulp reference materials are sourced from
haematite iron ores from the Pilbara region in Western
Australia and were certified using XRF in November 2007.
There were a total of 179 analyses each of GIOP-17 and
GIOP-19, with 120 analyses of GIOP-18. An assessment of
the reference material results, conducted as part of a review
of the project, found that the majority of results fell outside
acceptable control limits (defined by ±3 standard deviations
from the expected value) and that there was a significant
bias present. All three reference materials were reporting
approximately one per cent Fe below the expected grade
(Table 2). An example control chart for GIOP-18, showing the
strong negative bias, is provided in Figure 6.

FIG 4 - Q-Q plot showing bias in sample weights for field duplicate samples TABLE 2
collected using the four-tiered riffle splitter. Reference material results for Fe, showing bias in results from 2009 sampling
program.
splitter will cause a biased split when not levelled correctly,
especially along the long axis of the splitter, as the material Standard ID GIOP-17 GIOP-18 GIOP-19
will preferentially fall down the sides of the chutes rather than Certified value (% Fe) 58.5 60.6 63.5
the middle of the chutes. Consequently, the more tiers a riffle
splitter has, the less representative the final split is likely to be, Certified std dev (% Fe) 0.145 0.132 0.166
especially if the splitter is not levelled correctly. Count 179 120 179
In this case, the bias in field duplicate sample weights when Minimum (% Fe) 56.37 58.26 61.41
using the four-tiered riffle splitter was caused by a number of
Mean (% Fe) 57.48 59.65 62.49
factors, including:
Maximum (% Fe) 58.89 61.2 63.72
•• poor introduction of the sample into the top of the riffle
splitter (pouring from the bag); Difference (% Fe) -1.02 -0.95 -1.01
•• riffle splitter rigidly attached to trailer with no mechanism
Investigation into the source of the bias, including
to level the splitter, especially across the long axis of the
discussions with the laboratory, found that the bias was
splitter; and
caused by the combination of:
•• too many tiers, resulting in increment delimitation errors.
•• A drift correction error, identified by the laboratory, in the
The bias caused by the discrepancy in sample weights and XRF machine used during analysis.
poor splitting practices was not identified until after the •• The reference materials were not dried prior to analysis,
drilling had been completed, despite the data being collected. leading to higher than expected loss on ignition (LOI)
The result of which was a lower confidence in the reliability of results, which in turn resulted in underestimation of the
the sample data collected using the four-tiered riffle splitter. Fe content. Measurements of the moisture content of the

FIG 5 - (A) Trailer used during the 2006 to 2007 RC sampling program; (B) detailed view of the four-tiered riffle splitter mounted on trailer.

SAMPLING CONFERENCE / PERTH, WA, 11 - 12 MAY 2010 115

 J GRAINDORGE

FIG 6 - Control chart for Fe for reference material GIOP-18.

reference materials by the laboratory showed that the reliability of the data, rather than a proactive approach to the
moisture level was up to 0.46 per cent. analysis, whereby the errors are identified and corrected as
The net result of these errors was that the Fe content of they occur. This generally stems from a lack of understanding
the sample batches was consistently underestimated by regarding the basic principles of QAQC and the purpose of the
approximately one per cent Fe. Ultimately, the results were QAQC samples. A good QAQC system resolves the errors as
corrected by the laboratory involved. they arise and attempts to prevent the errors from occurring
Given the fact that the majority of results fell outside in future sample batches.
acceptable control limits along with the obvious bias present,
suggests that, while QAQC data was collected, it was not REFERENCES
adequately analysed and the purpose of the reference materials Carswell J T, Yulia K, Lesmana D and Steamy K, 2009. Grade
poorly understood. Had the QAQC data been regularly control sampling quality assurance/quality control in a high-
reviewed the bias would have been identified early on in the grade gold mine – Gosowong, Indonesia, in Proceedings Seventh
sampling program and potentially saved a lot of effort and International Mining Geology Conference, pp 283-290 (The
cost associated with fixing the errors. Studies show that the Australasian Institute of Mining and Metallurgy: Melbourne).
cost of fixing an error can be many times that of the original Greig, D and Cook, W, 1998. Methods for assay database validation,
work (Noppé, 2004). Upon receipt of assay results from the more meaningful sampling in the mining industry, The Australian
laboratory, and prior to their inclusion into the drillhole/ Institute of Geoscientists Bulletin, 22:31-37.
sample database, the results should be reviewed with the aim
Gy, P, 1998. Sampling for Analytical Purposes, p 153 (John Wiley &
of identifying obvious errors.
Sons Ltd: England).
Moreover, the investigation showed that the laboratory
JORC, 2004. Australasian Code for Reporting of Mineral
involved was treating the reference materials separately and
Resources and Ore Reserves, Joint Ore Reserves Committee of
differently to the routine samples, which is contrary to the
the Australasian Institute of Mining and Metallurgy, Australian
principle of using reference materials to monitor analytical Institute of Geoscientists and Minerals Council of Australia.
accuracy during routine assaying. In this case, the laboratory
admitted to removing the reference materials, which were Leppard, G, 1998. The direction of mineral laboratories and the
in foil sachets, from the sample batches. These reference emerging need to be more accountable and transparent, More
materials were then analysed as a separate batch, rather than meaningful sampling in the mining industry, The Australian
Institute of Geoscientists Bulletin, 22:47-53.
as part of the original sample batch. The preparation of the
reference materials was also different to the routine samples Noppé, M A, 2004. Reconciliation: Importance of good sampling and
as the reference materials were not dried prior to XRF analysis. data quality assurance-quality control (QA/QC), in Proceedings
EGRU Mining and Resource Geology Symposium, EGRU
Contribution 62, pp 107-113 (James Cook University: Townsville).
CONCLUSIONS
Pitard, F F, 1993. Pierre Gy’s Sampling Theory and Sampling
The three case studies provided highlight the need for QAQC
Practice, p 488 (CRC Press: Boca Raton).
data to be analysed in a timely manner. In all three cases, the
data shows a clear bias. Had this data been analysed during Simon, A and Gosson, G, 2008. Considerations on quality assurance/
quality control and sample security, in Proceedings Sampling
the drilling programs, rather than after, the errors would have
2008, pp 135-140 (The Australasian Institute of Mining and
been identified early on and corrected, providing significant
Metallurgy: Melbourne).
time and cost savings, and improving confidence in the
reliability of the sample data. Voortman, A G W, 1998. Avoiding over- and underestimation of
grade by using correct sampling, More meaningful sampling in
In the author’s experience, this is a common occurrence, the mining industry, The Australian Institute of Geoscientists
where good quality QAQC data is collected; however, the Bulletin, 22:21-29.
analysis of the data is only completed after the sampling
program has been completed. At this stage, the analysis
becomes merely an exercise in documenting the errors and

116 SAMPLING CONFERENCE / PERTH, WA, 11 - 12 MAY 2010