HOME

Some Key Facts and Concepts in the Evolution of Sampling and

Assaying Practices at Codelco
P Carrasco1
ABSTRACT
Incorrect sampling and measurement operations and misunderstanding
the components of variability can cause huge economic losses to the
mining industry. The main objective of this paper is to illustrate by means
of four historical facts how Codelco has been evolving in those issues
since the late 1980s. The El Abra bias, the awareness of the different kind
of sampling errors and the role of Francis Pitard, the awareness of
weighting errors, and the understanding of the components of nature and
process variability are described. It has been a slow but a consistent
process. Much progress has been achieved but still there are many
opportunities for improvement to transit from the tonnage culture to the
productivity culture.

At the same time Codelco was studying the feasibility of

mining the oxide zone of Radomiro Tomic, a world class high
tonnage-low grade porphyry copper. After the El Abra assaying
bias experience Radomiro Tomic copper assays were checked.
Unfortunately, an analytical bias was also discovered here.
Figure 1 shows the time evolution of the analytical error and
the analytical techniques at the laboratory involved in the bias,
based on pulp duplicates from Radomiro Tomic.

INTRODUCTION
In order to optimise the management of the mining business it is
very relevant to do good measurements. Unfortunately this is not
always the case in our mines. By good measurements I mean
accurate and precise or reproducible measurements. Accuracy
and precision are functions of the objectives of the decisions to
be made along the mining business value chain. Generally the
objective is to minimise the economical losses, not the hidden, in
a sort of individualistic way. As a consequence the optimisation
of the system is not achieved.
Some of the key variables to measure are grades and tonnages.
If the key measurements are not properly done, there is no
possible way to learn from the process and therefore continuous
improvement becomes a myth. The abundant models used to
design, plan and materialise the business cannot be verified and
validated and production cannot be reconciled.
Another of our main weaknesses is the lack of understanding
about the variability and more precisely the different kinds of
variability. This impedes us from optimising the net present
value because we are not able to understand the real nature of our
process from rock to cathodes, hence the hidden losses are
difficult to discover.
The main objective of this paper is to illustrate by means of
four historical facts how Codelco has been evolving in those
issues and to share with the sampling community some key facts
and concepts learned over the last 30 years. It has been a slow
but a consistent process. Much progress has been achieved but
still there are many opportunities for improvement to transit from
the tonnage culture to the productivity culture.

FACT ONE EL ABRA AND RADOMIRO TOMIC

ASSAY BIAS
In the late 1980s Codelco top management decided to sell part of
some mining assets. As a consequence 51 per cent of El Abra
was sold. During the due diligence process an analytical bias was
discovered. This bias indicated an overestimation of the mean
grade of the deposit of 0.05 per cent Cu in the oxide zone. The
bias was not constant; it was conditional to grade. As the grade
decreased the bias increased. For a high tonnage-low grade
deposit this bias was very significant from the economic view
point (Carrasco, Carrasco and Jara, 2004).

1.

Director de Especialidad Tecnica, Codelco, Chile.

Email: [email protected]

Sampling Conference

FIG 1 - Evolution of the analytical error over time and the analytical
techniques used at the laboratory.

This figure gives us many lessons:

Precision and accuracy are very variable over time. As a

consequence a systematic QA/QC program must always
exist.

Precision is a function of the technique. In this case AA was

always more precise than volumetry.

XRF is very imprecise and highly biased.

Atomic absorption is always more precise than the other
techniques. Late AA, although biased, is more precise than
the early AA.
Before 1974 the technique was short volumetry; as can be seen
it was biased, probably because of bad assaying practices. Then
AA equipment was acquired in order to modernise the assaying
process. At the beginning AA was accurate and then because the
AA was not matching the volumetry checks the AA results were
corrected. This was the main cause of the El Abra bias. Then
XRF was implemented by the chief chemist in order to increase
productivity. At that time the main mined ore was secondary
sulfides, therefore the chemist set the technique up accordingly.
The geologists did not explain they were sending oxides samples
from a different ore unit; the chemist did not explain to the
geologists the sensitivity of the technique to the matrix. This was
the main cause of the Radomiro Tomic bias.
Summarising, imprecision and inaccuracy could be explained
by several causes:

lack of QA/QC systems,

lack of communication between geologists and chemists,

Perth, WA, 27 - 29 May 2008

P CARRASCO

wrong optimisations strategies each one made honest efforts

to optimise their job ignoring the optimisation of the system in
a holistic way,

incompetence, and/or
ignorance.

incorporation of all the former issues into the Corporate

Capital System for new relevant projects (Cnam 016, 2004);

introduction of chronostatistics in order to better understand

the components of the variability of process;

The El Abra and Radomiro Tomic biases made top management

aware about the economical relevance of improper analytical
practices. As a consequence they lead a complete program of
QA/QC of chemical analysis and sampling.
Relevant actions were:

creation of a special group devoted to develop QA/QC at

corporate level;

acquisition of a chemical laboratory managed by very

experienced chemists in order to develop primary analytical
methods for SRM in our experience, after several
international round robin exercises, the only safe laboratory
is the one under your control;

development and implementation of corporate guidelines to

prepare standard reference materials from Codelco ores,
concentrates and tails (Cnam 009, 1998);

development and implementation of corporate sampling

guidelines for chemical analysis purposes (Corporate
Committee of Chemical Analysis, 1994);

development and implementation of corporate guidelines for

chemical analysis verification (Cnam 010, 2007);

systematic training courses; and

systematic audits.
FACT TWO AWARENESS OF SAMPLING
ERRORS AND THE ROLE OF FRANCIS PITARD

introduction of the concept of chronostatistics process control;

improvement of weighing systems (Cnam 014, 2004); and
systematic audits.
FACT THREE AWARENESS OF
WEIGHING ERRORS
The proper measurement of the tonnage is very relevant for the
mining industry. The estimation error of the in situ tonnage is
relevant to determine the incertitude classification of the ore
resources (inferred, indicated or measured). Many challenges for
the geologist exist in order to solve this problem, because the
incertitude of the in situ tonnage is a function of the incertitude
of the volume of the geological unit that hosts mineralisation, the
incertitude of the specific gravity and the incertitude of the
moisture content.
The mining tonnages are also an issue. A study proved the
mine tonnages biased when estimated by truck constant factor.
As a matter of fact the mine tonnages were overestimated by
five per cent. Another study proves the weightometers biased at
several belt points in the plants. The lesson was: generally, when
the weightometers are not properly calibrated the tonnages are
overestimated. The magnitude of the overestimation varied
between four per cent and 14 per cent.
The key experiment to be convinced about the unreliability of
improper calibrated belt scales was the measurement of a big
tonnage in a very precise scale (material test). Figure 2 to Figure 6
show the measurement process (Wilke, 2007).

Another relevant action trigged by the El Abra bias was to hire

the services of Francis Pitard, a world reknown sampling
consultant, disciple of Pierre Gy and Ingamells.
In the early 1980s our knowledge of the sampling theory was
limited. Indeed, we were just aware of the fundamental error. We
made some efforts to better understand the theory but
unfortunately the existing books were very cryptic. Francis
Pitard played a fundamental role in educating us about the theory
of sampling. As well, he helped us to convince top managers to
improve sampling and assaying systems all along the mining
value chain from rock to cathodes. Also, he encouraged us to
better understand the relevance of process variability to optimise
the mining business and to begin the transition from the culture
of tonnage (quantity) to the culture of productivity (quality).
Some major improvements at corporate level were:

development and implementation of guidelines to estimate

FIG 2 - Deviating the material.

the sampling constants of Codelco mines ore types (Cnam

008, 1998; Cnam 011, 2000; Carrasco et al, 2005);

development and implementation of guidelines for the

optimisation of sampling protocols for ores, concentrates, tails,
anodes, cathodes, etc (Cnam 013, 2003; Cnam 015, 2004);

Some major improvements at corporate level were:

research and training program with the Geosciences Center of

development and implementation of guidelines to improve

the Paris School of Mines to apply the truncated Gaussian

simulation method to the assessment of the volume incertitude
of ore units (Carrasco et al, 2007);

the quality of sampling preparation equipment (Cnam 012;

Cnam 013, 2003);

improvement of volume and mass measurements at several

systematic courses of sampling theory for operators,

development and implementation of guidelines to improve the

geologists, mining engineers, metallurgists and chemists;

systematic reports for top managers;

design and building of head, concentrate and tails sampling
stations in all Codelco concentrators and projects (Cnam 007,
1996);

scales (drill core, mine bench, etc);

quality of weighing systems (Cnam 014; Cnam 016, 2004);

development and implementation of systematic material tests;

and

an ongoing research program to implement the measurement

of mass by proportional sampling (Gy, 1998).

Perth, WA, 27 - 29 May 2008

Sampling Conference

SOME KEY FACTS AND CONCEPTS IN THE EVOLUTION OF SAMPLING AND ASSAYING PRACTICES AT CODELCO

FIG 5 - Loading.

FIG 3 - Stockpiling.

FIG 6 - Weighing.

The science which studies the variations in the space and/or

time is called geostatistics. It was developed by Georges
Matheron at the Paris School of Mines in the early 1960s
(Matheron, 1962). Since then, geostatistics has made significant
progress because of Matheron and the contributions of his
disciples in France and around the world.
Some geostatistical concepts are essential to understanding
natural and process variability:

Support effect the perception of variability is a function of

FIG 4 - Final stockpile.

FACT FOUR UNDERSTANDING VARIABILITY

The understanding of variability and its components always
conducts to economical optimisation of the mining business.
Unfortunately, one of the main weaknesses is the lack of
understanding of the variability from rock to cathodes. Indeed,
the top management generally makes decisions by using
averages. Averages summarise too much in such a way that
crucial information is lost. More interesting is to learn from the
natural and mining process by studying the variations along
space and time. As Louis D Brandeis said:
I abhor averages. I like the individual case. A
man may have six meals one day and none the
next, making an average of three meals per day,
but that is not a good way to live.

Sampling Conference

the volume of the samples. As the sample volume increases

the variance of the random function decreases. Table 1 shows
the gold grades of diamond drill holes and their neighbouring
blastholes.
TABLE 1
The gold grades of diamond drill holes (DDH) and their
neighbouring blastholes.

Mean (g/t)
2

Variance (g/t)

Blastholes

DDH

1.59

0.87

9.23

5.21

The first conclusion from the table is the existence of a

contradiction. The variance of the blasthole grades cannot be
higher than the variance of the diamond drill holes because the
sample volume of the blastholes is by large bigger than the

Perth, WA, 27 - 29 May 2008

P CARRASCO

sample volume of the diamond holes. Besides, the mean grade of

the blastholes is significantly higher of the mean grade of the
diamond drills. The conclusion is that the diamond drill samples
are not capable to represent the variability of the gold grades
because its volume is too small and the sample number
insufficient. This is possible in high nugget effect skewed
lognormal populations where the mean is higher than the mode.

Information effect the perception of variability is a function

of the amount of information and the quality of the
estimation. As estimation is never equal to reality, decisions
made on estimates could be very different to decisions made
on true values. This is very important in selective mining of
complex ores.

Figure 8 is explicit about the variations inside a half shift

period of time and about the uselessness of the average. The
plant manager would be very happy with the performance if only
the average is seen. A very different perception arises when
looking at the grades on a 6 min basis. This study allowed
finding out the origin of the low-grade values. The cause was the
increase of ore coming from draw points with high dilution
content. As a consequence mine planning and drawing practices
were improved.

Variability and scale the variability is a function of the

scale. Figure 7 shows the variance as a function of the scale
for a stationary random function at different ranges.

FIG 8 - Variability of the copper head grade versus time.

FIG 7 - Variance as a function of the scale for a stationary random

function at different ranges.

As many wise managers know, the understanding of variability

is always translated into better economical results (Carrasco,
Carrasco and Jara, 1994).
The following example is explicit. Let us consider several
cyclic random functions with the same mean grade but different
variances for the copper grade. Also let us assume that the low
grade ore has a lower recovery than the high grades. Figure 9 and
Figure 10 show the results.

Sampling and assay errors if the sampling errors are

independent of the grade, mutually independent and of the
same variance, then the error variance adds to variogram as a
nugget component (Chiles and Delfiner, 1999). This additional
variability does not belong to the nature nor to the process. It
is called irrelevant variability. Misunderstanding the irrelevant
variability can be very expensive (Carrasco, Carrasco and Jara,
2004; Carrasco, Wilke, Jara and Suarez, 2007).

Variogram the variogram is a powerful tool to study natural

and process variability. Many messages are inside the
experimental variograms. It is very useful to study continuity,
range of influence, stationarity and anisotropy, and is
essential for the application of all geostatistical estimation
and simulation techniques.
Maybe because of cultural reasons the understanding of
variability is uncommon. As a matter of fact, the Judean-Christian
culture is essentially deterministic. In addition the occidental
science has an important deterministic heritage. As a consequence,
to convince about the economical importance of variability and to
fight against the very common paradigm of homogeneity is not an
easy task. In our experience one crucial experiment was very
important to convince the top management about the importance
of variability. The variation of the cooper head grade was
measured at a frequency of 6 min for 4 h. Figure 8 shows the
results.

FIG 9 - Head copper grade versus time.

Surprisingly, the recovery is very sensible to the variations of

the head grade variability in such a way that when the grade
variability decreases the metallurgical recovery increases.

Perth, WA, 27 - 29 May 2008

Sampling Conference

SOME KEY FACTS AND CONCEPTS IN THE EVOLUTION OF SAMPLING AND ASSAYING PRACTICES AT CODELCO

chief at El Teniente, Victorino Moyano, former QA/QC at

Codelco Norte, Eduardo Jara, Principal Geostatistician and
Everardo Suarez, Trainee Engineer at central office, for their
commitment to continuous improvement of sampling, assaying,
material balance and process control practices and for their
precious collaboration.

REFERENCES

FIG 10 - Metallurgical recovery of copper versus variability.

Some major improvements at corporate level were:

development and implementation of guidelines to consider

the variability in ore resource estimation and grade control;

systematic courses of statistics, linear geostatistics,

non-linear geostatistics and chronostatistics for geologists,

mining engineers, metallurgists and chemists; and

research work to quantify the economical impact of natural

and process variability misunderstanding.

CONCLUDING REMARKS
Improper sampling, assaying and weighing practices and the
misunderstanding of the natural and process components of
variability can produce monumental value losses to the mining
industry worldwide. Those losses lead as well to economic
inefficiency and unsustainable exploitation of natural resources,
jeopardising the wealth of future generations and adding
unnecessary negative externalities to society. The mining
industry has a magnificent opportunity to increase their
economic performance by discovering hidden losses. This can be
done by applying optimal assaying techniques, the principles of
the sampling theory, statistical and geostatistical thinking,
effective chronostatistical process control, minimising the
irrelevant variability, maximising the understanding of relevant
process variability and by encouraging the work of
multidisciplinary high level experts aligned with the main
objectives of the mining business.

ACKNOWLEDGEMENTS
The author wishes to thank Mr Fernando Vivanco, corporate
Projects Vice President, and Mr Julio Beniscelli, Manager of
Technical Services, Codelco Chile, for their strong support in the
implementation of good sampling, assaying and geostatistical
practices along the copper business value chain.
I am also grateful to Francis Pitard for his wise advice and
lectures on sampling, assaying, chronostatistics and process
control practices over the last 15 years.
I also would like to thank Patricio Guerra, Chief Chemist at El
Teniente, Julio Tapia, Magali Campos and Ester Menichetti,
chemist consultants at central office, Alfredo Wilke, QA/QC

Sampling Conference

Carrasco, P, 1998. Norma Codelco Cnam 008: Test de heterogeneidad para

determinar constantes de muestreo y nomogramas de preparacin de
muestras, Codelco Chile, Casa Matriz, informe interno (unpublished).
Carrasco, P, Campos, M, Tapia, J and Menichetti, E, 2000. Norma Cnam
011: Test de Ingamells, Codelco Chile, Casa Matriz (unpublished
report).
Carrasco, P, Campos, M, Tapia, J and Menichetti, E, 2003a. Norma Cnam
012: Operaciones estndar a seguir durante el sub muestreo en el
laboratorio, Codelco Chile, Casa Matriz (unpublished report).
Carrasco, P, Campos, M, Tapia, J and Menichetti, E, 2003b. Norma Cnam
013: Aseguramiento de calidad en preparacin de muestras minerales
para analisis quimico, Codelco Chile, Casa Matriz (unpublished
report).
Carrasco, P, Carrasco, P, Campos, M, Tapia, J and Menichetti, E, 1998.
Norma Cnam 009: Obtencin y certificacin de materiales de
referencia en los laboratorios qumicos, Codelco Chile, Casa Matriz
(unpublished report).
Carrasco, P, Carrasco, P, Campos, M, Tapia, J and Menichetti, E, 1999.
Norma Cnam 010: Verificacin de resultados de anlisis qumicos,
Codelco Chile, Casa Matriz (unpublished report).
Carrasco, P, Carrasco, P, Campos, M, Tapia, J and Menichetti, E, 2005.
Heterogeneity and Ingamellss tests of some Chilean porphyry ores,
in Proceedings Second World Conference on Sampling and Blending,
pp 139-150 (The Australasian Institute of Mining and Metallurgy:
Melbourne).
Carrasco, P, Carrasco, P, Ibarra, F, Rojas, R, Le Loch, G and Seguret, S,
2007. Application of the truncated Gaussian simulation method to a
porphyry copper deposit in Proceedings APCOM, 33rd International
Symposium on Application of Computers and Operations Research in
the Mineral Industry, pp 31-39 (University of Chile: Santiago).
Carrasco, P, Carrasco, P and Jara, E, 2004. The economic impact of
correct sampling and analysis practices in the copper industry,
Chemometrics and Intelligent Laboratory Systems, 74:209-213.
Carrasco, P and Pitard, F, 2004. Norma Cnam 015: Muestreo y analisis de
ctodos de cobre, Codelco Chile, Casa Matriz (unpublished report).
Carrasco, P and Tello, A, 1996. Norma Cnam 007: Condiciones de
estaciones de muestreo de flujos de materiales y transferencia de
productos, Codelco Chile, Casa Matriz (unpublished report).
Carrasco, P, Wilke, A, Jara, E and Suarez, E, 2007. Some applications of
statistics and geostatistics to the sampling of a big cooper tail stream,
in Proceedings Third World Conference on Sampling and Blending,
pp 352-366.
Chiles, J P and Delfiner, P 1999. Geostatistics: Modelling Spatial
Uncertainty, 695 p (John Wiley and Sons, Inc).
Corporate Committee of Chemical Analysis, 1994. Normas para analisis
quimico, segunda edicin, Codelco Chile, Casa Matriz (unpublished
report).
Gy, P, 1998. Sampling for Analytical Purposes, pp 117-118 (John Wiley
and Sons, Inc).
Matheron, G, 1962. Trait de gostatistique applique. Tome 1, 334 p;
Tome 2, 172 p (Editions Technip: Paris)
Solis, O, Wilke, A, Carvajal, P and Tapia, J, 2004. Norma Cnam 014:
Sistemas de pesaje en correas transportadoras, Codelco Chile, Casa
Matriz (unpublished report).
Wilke, A, 2007. Mejoramientos al balance metalrgico, Codelco,
Divisin El Teniente (unpublished report).
Wilke, A, Solis, O and Ans, J, 2004. Norma Cnam 016: Antecedentes
metrolgicos para proyectos, Codelco Chile, Casa Matriz (unpublished
report).

Perth, WA, 27 - 29 May 2008