Paper 30:text

10/10/09

9:04 AM

Page 235

MARQUES, D.M., COSTA, J.F.C.L., RIBEIRO, D.T. and KOPPE, J.C. The evidence of volume variance relationship in blending and homogenization piles
using stochastic simulations. Fourth World Conference on Sampling & Blending, The Southern African Institute of Mining and Metallurgy, 2009.

The evidence of volume variance relationship in blending and

homogenization piles using stochastic simulations
D.M. MARQUES*, J.F.C.L. COSTA*, D.T. RIBEIRO and J.C. KOPPE*
*Mining Engineering Department; Federal University of Rio Grande do Sul
DSc; Vale

Homogenization piles are largely used in the mining industry for variability reduction in the head
grades from the ores feeding the processing plants. Variability reduction in large piles is based on
the principle volume-variance relationship, i.e. the larger is the support the smaller is the
variability. The methodology proposed combines chevron type piles and geostatistical simulation
to emulate the in situ and the pile reclaimed grade variability. Based on a pre-defined mining
sequence to select the blocks that will form each pile for each simulated block model, the
statistical fluctuation of the grades derived from real piles can be simulated, at different pile sizes
and number of layers used in pile construction. Using this methodology, one can evaluate within a
certain time period the expected grade variability for various pile size and, also, the internal grade
variability when a given pile is reclaimed. The results forecasted by the method were compared
against the real grades obtained by the grade control system used at a large iron mine operated by
Vale during 2004.
Keywords: piles, homogenization, variability reduction, iron ore.

Introduction
Ore recovery at mineral processing plants can be affected
by head grade variability. Generally, ore recovery is
maximized when head grades, from the ore feeding the
plant, are kept within narrow intervals of variation. The
reduction in this interval of grade feed variability can be
reached via adequate mining scheduling, finding the most
feasible stationary mining route and/or homogenizing ore
using large blending piles.
Ore heterogeneity can be evaluated on several scales. The
types of heterogeneities found during ore preparation are
related, essentially, with the geological processes involved
in the mineral deposit genesis and with the equipment and
methods used in mining, handling and transporting this ore.
According to Schofield (1980)1, shipment, transportation,
comminution, storage and handling contribute in some way
to the variability found in the grades, since they introduce
certain amount of reverse structuring on the space
distribution of the natural mineralization.
Parker (1979) 2 shows the intrinsic relation between
support (volume) of a lot and grade variability. The
quantification of volume variance relationship demands the
knowledge of mineralization methods and the use of proper
geostatistical methods.
Most estimation techniques provide a block model with
grades assigned to each block using the methods of kriging.
(Matheron, 1963)3. This block model is inappropriate for
accessing the uncertainty associated with the estimated
grades. Consequently, this procedure fails in predicting
grade fluctuations correctly. Geostatistical simulations aim
rather, at reproducing in situ variability, provided that the
spatial continuity of the input data set is respected. The
simulated model is said to be conditionally simulated, if it
honors the values that are obtained at sampled points while

reproducing the same dispersion characteristics of the

original data set, (i.e. the mean, variance and covariance or
variogram functions). In a conditionally simulated model,
the dispersion of the grades during mining or processing
can be addressed, since the dispersion characteristics of the
original data are maintained. As the spatial continuity and
variability of the real deposit is properly determined,
quality reliability in the numerically simulated model is
obtained, helping to assess potential risks during the
decision making process.
Some previous studies involving blending and
homogenization piles and geostatistical simulations are
described by Schofield (1980) 1 , Costa et al. (2007) 4 ,
Ribeiro et al. (2008)5.
This paper presents a study developed at two iron ore
deposit, in southeast Brazil. It investigates the in situ grade
variability obtained during mining and the influence of pile
dimension to attenuate this variability. SiO2 was chosen as
the main variable for this analysis as it constitutes the most
critical and variable contaminant present in the ore.

Methodology
Before starting the analysis on the variability reduction with
the increase of the pile mass, it is necessary to understand
all variability sources influencing the system, namely:
Intrinsic variability of the ore grades from Mine 1
Intrinsic variability of the ore grades from Mine 2
Variability of the grades, which feed the
homogenization system (pre-pile combining ore from
several mining benches from the two mines)
Variability among grades coming from different
homogenization piles (variability inter-piles depending
on pile mass).
Figure 1 displays a schematic representation of the

THE EVIDENCE OF VOLUME-VARIANCE RELATIONSHIP IN BLENDING

235

10/10/09

9:04 AM

Page 236

SiO2 (%)

SiO2 (%)

SiO2 (%)

SiO2 (%)

Paper 30:text

Figure 1. Diagram showing a typical blending system used at some Brazilian iron mines. Black dots represent the locations where samples
were obtained

each blending stage.

In situ grade variability is obtained using diamond drill
hole core assays from mine 1 and 2 and also with the
following steps:
generation of multiple equally probable 3D models via
sequential Gaussian simulation (sGsim)
determination of optimal scheduling for the 3D block
model input
calculation of the variability in each stage for every 3D
simulated block model,
The simulated block models are required to be validated
prior to their use in defining uncertainty limits. Validation
of stochastic models comprises visual checking to verify the
correct reproduction of the spatial continuity and data
conditioning, and histogram and variogram reproduction
and their ergodic fluctuations.
A 3D grade block model was built using sequential
Gaussian simulation (sGsim) (Isaaks, 1990) 6 . This
algorithm generates multiple and equally probably
scenarios for the grade spatial distribution, reproducing
statistically, its spatial continuity and histogram.
Simulated grades at various supports provide the means
to forecast pile grade fluctuations. These fluctuations can be
plotted as confidence intervals. If the process is valid the
real grades determined by the sampling system should fall
within the interval predicted by the simulations.

Variance

analyzed system. The black dots in this figure highlight the

sampling locations along the circuit, also the variability
plots at those locations are also depicted. These plots will
be further discussed.
The sampling is carried out along the conveyor belt at
every 600 t interval. These materials are homogenized at
the end of the shift and sent for laboratory analysis.
Each block mined has a different grade. These grades differ
among them and they are also different from the global mean
grade mined along the first production year. By grouping
various blocks and forming a pile, the average grade of this
pile is closer to the global annual mean if compared to the
grades of each individual block which comprise the pile. This
phenomenon of variance reduction (Figure 2) is well known
among geostatisticians and it is referred to as the volumevariance relationship (Parker, 1979)2.
The data collected at the mine site includes production
information at every stage along the homogenization
system within the year of 2004. The smallest scale of
available information corresponds to a production shift,
including the tonnes produced and their grades. The
analysed variable is silica (SiO2), an erratic component in
this type of mineral deposit.
The data and information acquired, made it possible to
proceed with a comparative analysis on the variability in
the grades derived from the two mines and the effect along

Figure 2. Volume-variance relationship. As the mass of the lot analyzed increases (mass of the pile) (a), the variance among the grades
reduces (b)

236

FOURTH WORLD CONFERENCE ON SAMPLING & BLENDING

Paper 30:text

10/10/09

9:04 AM

Page 237

Results and discussion

Assessing block grade uncertainty
Data available include vertical drill holes, comprising 5 928
composites from mine 1 and 3 585 composites from mine 2,
which have their support regularized at 10 m intervals.
Conditional geostatistical simulations (sGsim) were used
to obtain 50 equi-probable scenarios for the attribute of
interest at the block size associated to the mining selectivity
to be adopted (10 m 10 m 10 m).
Mine sequence
Reconciliation results for grade blocks were not available at
a block by block basis. The mines, during each month were
mined in polygons and these polygons include all blocks
extracted in a given period without any specific order.
Basically, the sequence of blocks extracted along a month is
not available. Consequently, we sorted out the blocks
within each given month (polygon) by x, y and z
coordinates.
Pre-product variability
The material analyzed before arriving at the processing
plant is the so-called Pre product, which is basically ROM
from mine 1 and 2. This material is sampled at the conveyor
belt which transports ore from each mine.

Mine 2
The lots analyzed at each shift weights 8.2 ktonnes on
average with a 3.2 ktonnes standard deviation, (Figure 5a).
The SiO2 average grade mined along 2004 is 2.64% with a
standard deviation of 1.42%. Grades from mine 2 are more
erratic than the ones from mine 1. Figure 4b presents the
time series for all grades from ore lots produced at mine 2
along 2004.
It can be noticed at Figure 3b and 5b, the large variability
from SiO2 produced along each shift at each mine. This
variability would lead to a yield reduction at the processing
plant with obvious operational problems.
It can be verified that practically all the sampled values
fall in the upper-lower uncertainty limits predicted by the
50 simulations. Note, that there is not a direct block to
block (block grades predicted by simulation and the block
sampled at the blending system) correspondence at plot 5
and 6. As it is not possible to identify within a given month
the exact sequence of blocks mined, probably the sequence
of blocks (scheduled) used for the simulations do not match
the real one adopted at the mine. This leads to a few
readings (real grades) plotting outside the predicted limits.
Pre pile variability
The pre pile material includes ROM from the two mines. It
is the combination of the ore from mine 1 and 2. The lots

SiO2 (%)

Frequency

Mine 1
The average mass of the lots analysed during a shift is 8.6
ktonnes with a standard deviation of 2 ktonnes (Figure 3a).
The SiO2 average grade mined along the year was 2.21%
with a standard deviation of 1.07%. Figure 3b presents the
time series for all grades from ore lots produced at mine 1
along 2004.

Figure 4 shows three series with SiO2 grades. Note, the

lower and upper bounds obtained by the 50 simulations and
also the measured (real) grades. Each point plotted
corresponds to approximately 8.7 ktonnes mass which
refers to three blocks.
It can be verified that practically all the sampled values
fall in the upper lower uncertainty limits predicted by the 50
simulations.

SiO2 (%)

Figure 3. (a) Histogram for the mass of the lots (b) SiO2 grades at each 8.6 ktonnes lot along 2004

Figure 4. SiO2 grades at each 8.6 ktonnes lot along 2004 (red dots), upper (green line) and lower (blue) bound for simulated SiO2 grades

THE EVIDENCE OF VOLUME-VARIANCE RELATIONSHIP IN BLENDING

237

10/10/09

9:04 AM

Page 238

SiO2 (%)

Frequency

Paper 30:text

SiO2 (%)

Figure 5. (a) Histogram for the mass of the lots (b) SiO2 grades at each 8.2 ktonnes lot along 2004

Frequency

SiO2 (%)

Figure 6. SiO2 grades at each 8.2 ktonnes lot along 2004 (red dots), upper (green line) and lower (blue line) bounds for SiO2 grades

Figure 7. (a) Histogram for the masses of the lots of pre-pile material (b) time series depicting SiO2 grades from the pre pile material (mine
1 + mine 2) along 2004

analyzed along each shift weighs approximately 19 ktonnes.

Figure 7 presents the masses of lots at the pre-pile stage and
their SiO2 grades.
Note in figure 7b, there is a significant reduction in
variability, if compared against the ones obtained from the
grades at each mine individually (Figure 3b and 5b).
In Figure 8, each value plotted corresponds to every three
re-blocked grades from mine 1 and 2, (i.e. approximately
17.4 ktonnes).
The average silica grade for the combined lots from the
two mines is 2.28%, and the standard deviation is 0.66%.
The variability of the initial grades was reduced by 40%
blending the ores from the two mines.
Inter piles variability
There are various constructive methods for pile
homogenization and among them the most common is the
chevron type. The chevron system at Vale's iron mines
contains two or a multiple of two long, rectangular beds.
While a pile is being built on a bed, the pile on the other

238

bed is being reclaimed (Figure 9). The company can set up

piles of up to 200 ktonnes each (max two piles).
Along the year 2004, 135 piles with different masses
were mounted (Figure 10a). These differences in pile
masses were caused by high market demands for iron ore;
this didnt allowing the company to stock material at piles.
The consequence was less mass at the piles than planned.
Figure 10b presents grades from each pile following the
order they were mounted (time series). The variability of
grade values on this plot (Figure 10b) is less than the
variability at the previous stage (Figure 7b).
As the 137 piles have different masses they were divided
into 27 classes of similar size with at least five piles in each
class. Within each class (which include piles of similar
size), the variability of the piles average grade was
calculated in relation to the average grade of the year 2004.
The process was applied to the 27 classes and the variability
for each of them calculated. The results obtained are plotted
at Figure 11, (i.e. the mass of the class versus the variability
for each pile size). Variability is expressed through the

FOURTH WORLD CONFERENCE ON SAMPLING & BLENDING

10/10/09

9:04 AM

Page 239

SiO2 (%)

Paper 30:text

Figure 8. SiO2 grades at each 19 ktonnes lot along 2004 (red dots), upper (green line) and lower (blue line) bounds for SiO2 grades

SiO2 (%)

Frequency

Figure 9. (a) Chevron type pile. (b) Sketch presenting the homogenization scheme used at Vales iron ore mines

Figure 10. (a) Histogram for the mass of piles formed along 2004 (b) SiO2 grades for the 137 piles formed within the year

coefficient of variation. It is demonstrated (Figure 11) that

there is a general trend in decreasing the variability as the
pile size increases. This plot also illustrates the principle
embedded in the volume variance relationship and explains
why the reclaimed grades after the blending piles have a
lower variability than the grades from the blocks leaving
the mine.
Figure 12 depicts the upper and lower bounds for the
expected grades for each pile mass obtained by sGsim and
the chevron pile emulator. Each 3D simulated block model
was used to construct multiple piles which will be formed
along a scheduled mining period (month). There are 50
simulated models each with different grade values for a
given block, leading to different grades for the same pile.
The piles analyzed are equivalent to 50, 75, 100, 125, 150,
175, 200, 225 and 250 ktonnes. As the block grade
variability is incorporated, the uncertainty limits can be
evaluated and compared against the real pile grades along
the year.
Note that the larger the pile, the better will be the
blending process; however, from an operational perspective
the problems and costs tend to increase as the size of the

equipment involved increases. The adequate pile size is the

minimum size which will deliver ore to the plant with
grades varying within a pre determined and acceptable
grade interval.
Composition of the variability of the homogenization
system
Variability was measured along different stages within the
production chain (Table I). It started by measuring ore
grades variability from the ore leaving each mine which is
basically the variability from ROM ore grades. The
combination of ROM from the two mines leads to the socalled pre pile variability. The last stage is the variability
after the blending piles which were tested for various pile
sizes. All these readings were combined in a single plot
(Figure 12a) depicting the amount of variability as a
function of pile mass or support (volume) of the sample. A
model was adjusted for the experimental results leading to a
volume variability curve. It can be read along this model
the expected coefficient of variation for SiO2 for any pile
size or support.

THE EVIDENCE OF VOLUME-VARIANCE RELATIONSHIP IN BLENDING

239

10/10/09

9:04 AM

Page 240

CV (%)

Paper 30:text

SiO2 (%)

Figure 11. Reduction on grades variability (measured by the coefficient of variation - CV) versus the mass of the piles

CV (%)

CV (%)

Figure 12. SiO2 grades at each pile along 2004 (correspond to real grades, red dots), upper (green line) and lower (blue line) bounds for
SiO2 grades obtained by sGsim and chevron pile emulator

Figure 13. (a) Variability as a function of mass, including the readings after mine 1 and 2 and after combining the ore from the two mines
(pre-pile) (b) model adjusted to the decay in variability as the pile mass increases

Table I
Comparative among SiO2 grades at different sampling stages
Sampling
point
Mine 1
Mine 2
Pre-pile
Piles

240

Average
mass
(ktonnes)
8.6
8.2
19
150

SiO2
CV (%)

Mean (%)

Min(%)

Max(%)

0.49
0.54
0.29

2.21
2.64
2.28

0.66
0.63
0.84

12.08
18.74
7.35

0.20

2.31

1.40

4.04

Conclusions
Silica variability was reduced to 40% of the initial grade's
variability by mining multiple benches and operating two
mines simultaneously. After the homogenization piles, it
was verified that it is possible to reduce in up to 75% on the
initial silica grades variability. The decay on grades
variability follows an exponential law as a function of the
mass of the pile. This corroborates the volume-variance
relationship.
The uncertainty limits can be properly forecast using
geostatistical simulation. The limits predicted included the

FOURTH WORLD CONFERENCE ON SAMPLING & BLENDING

Paper 30:text

10/10/09

9:04 AM

Page 241

real grades obtained by sampling along various stages from

the blending system.

Acknowledgements
CNPq is acknowledged for supporting scholarships to
students associated with this project. Vale (Companhia Vale
do Rio Doce) is acknowledged for the continuous support
the mining engineering research team at Federal University
of Rio Grande do Sul.

References
SCHOFIELD, C.G. Homogenisation/Blending Systems
Design and Control for Minerals Processing,
TransTech Publications. 1980.
PARKER, H. The Volume Variance Relationship: A Useful
Tool for Mine Planning. Engineering and Mining
Journal, vol. 180, 1979. pp. 106123.
MATHERON, G. Principles of Geostatistics, Economic
Geology, vol. 58, 1963. pp. 12461266.

COSTA, J.F.L.C., KOPPE, J.C., MARQUES, D.M.,

COSTA, M.S.A., BATISTON, E.L., PILGER, G.G.
and RIBEIRO, D.T. Incorporating in situ Grade
Variability Into Blending Piles Design Using
Geostatistical Simulation, in Proceedings of the Third
World Conference on Sampling and Blending, 2007.
pp. 378389.
RIBEIRO, D.T., STANGLER, R.L., VIDIGAL, M.,
COSTA, J.F.L.C. and Marques, D.M. Conditional
Simulations To Predict Ore Variability And
Homogenization Pile Optimal Size: A Case Study Of
An Iron Deposit, in Proceedings of VIII International
Geostatistics Congress, vol. 2, 2008.
ISAAKS, E.H. The Application of Monte Carlo Methods to
the Analysis of Spatially Correlated Data, PhD
Thesis, Stanford University, USA. 1990.

Diego Machado Marques

Graduate student, LPM - UFRGS
Graduate in Mining Engineer at Universidade Federal do Rio Grande do Sul (2008). Experience in
geostatistics, resource estimation. Publications in the field of geostatistics.
Presently MSc candidate at UFRGS in geostatistics applied to improve ore blending.

THE EVIDENCE OF VOLUME-VARIANCE RELATIONSHIP IN BLENDING

241

Paper 30:text

242

10/10/09

9:04 AM

Page 242

FOURTH WORLD CONFERENCE ON SAMPLING & BLENDING