You are on page 1of 15

Review of non-additivity and

density sampling impact for linear


estimate of SPI
Orlando Rojas - Principal Geologist Consultant, GeoEstima, Chile
Javier Pizarro - Senior Metallurgical Consultant, Empírica Consultores, Chile
Introduction
• After a tough work of sampling, characterization, and metallurgical testing we
need to populate block models with geometallurgical parameters to allow the use
of this information in the mining planning process or any other decision stage
• Estimates of metallurgical parameters using linear interpolator as ordinary kriging
(OK) and based on metallurgical testing made on drillhole samples, is a useful and
more frequently used approach, to populate block models with geometallurgical
parameters.
• Nevertheless, applying classical linear averaging resource estimation techniques
may not be appropriate because many of the geometallurgical variables are
complex and non-additive.
• In simple terms, applying linear interpolators for no-additive variables can produce
under or overestimation of true response value (see Figure) (Coward et al 2009).
Potential bias: Jensen’s inequality

Bias?

Bias?

Note:
The magnitude of
the bias depend on
how convex o
concave is the no-
linear function
Estimates of SPI  SEC

SEC* = 0.0915 x SPI* - 0.7082


X1,SPI1 X2,SPI2
ʎ1 ʎ2
SPI*
SEC*

ʎ3 ʎ4

X3,SPI3 X4,SPI4

“Estimates of SPI in the block model using the SPI determined from metallurgical testing
on samples, allow us estimate the SEC in each block”
Objective: to assess the potential bias
and review information effect
• In this work, it has been created an exercise to assess how could the non-additivity
feature affect the accuracy of SPI™ parameter estimates, under a linear approach

• Also, a review of impact of the sampling density for SPI™ estimates has been
assessed.
SPI™ & SEC
• The SAG power index (SPI™) is a
metallurgical test developed by Minnovex
to determine the time (expressed in
minutes) to reduce ore from 100% under
3/4“ and 80% under ½” until a size 80%
under 10#Ty.
• This is a “response variable or secondary
variable”
• Using this test it is possible to determine
the Specific Energy Consumption (SEC)
through the following equation
(Minnovex), at industrial scale for a feed
The factors “a” and “b” are coefficient or model constants
with an F80 of 6” (152.4 mm), in circuit
with uncrushed pebbles being recirculated K80 is a transfer size from SAG to ball mill
to the SAG mill.
Methodology
• To assess the potential under or overestimation of SPI™ estimates using Ordinary
Kriging (OK) as linear technique, a set of realisations were created based on Non-
Conditional Simulation technique
• This process has been supported in a real data of a well known deposit of Chile
• One of the 100 realisations have been chosen as “reality”
• Then, six datasets of fictitious drill holes were created to obtain different grid
spacing sampling set
• In each one the values of the chosen realization were recorded.
• Based on each data set, estimation models of SPI™ were performed using OK,
thereby we have obtained different SPI™ estimates and,
• At the same time, a set of estimation models related to different spacing
• A reconciliation between the “real” SPI™ and the value estimated in the different
linear models allow us to compare average values and variability according
different grid spacing sampling set.
Fictitious drillholes
Grid spacing 60 x 60 m Grid spacing 100 x 100 m

Drillhole

Grid spacing 80 x 80 m Grid spacing 120 x 120 m

Volume =
500x500x100m
 64 Mton
(aprox)
True vs. Estimated Models
Estimation I Estimation III

“True” or Reality

Grid spacing 60 x 60 m Grid spacing 100 x 100 m

Estimation II Estimation IV

Grid spacing 80 x 80 m Grid spacing 120 x 120 m


Reconciliation by spacing drillholes:
estimated models vs. true values
Estimates using grid
spacing of 60x60m

• There is almost no
global bias
• Low conditional bias

Estimates using grid


spacing of 80x80m

• There is almost no
global bias
• Conditional bias
Reconciliation by spacing drillholes:
estimated models vs. true values
Estimates using grid
spacing of 100x100m
• There is almost no
global bias
• Hight conditional bias

Estimates using grid


spacing of 120x120m
• There is almost no
global bias
• Very High Conditional
bias
Global bias for SPI is not relevant
SPI estimates statistic by spacing grid
Grid spacing N° Blocks Minimum Maximum Mean SD Difference Bias
spi_24_60x60 4012 20.12 59.43 42.16 52.18 0.3 0.72%
spi_24_80x80 4221 18.37 57.96 42.25 44.54 0.39 0.93%
spi_24_100x100 4367 26.06 58.27 42.02 33.2 0.16 0.38%
spi_24_120x120 4375 22.04 55.4 42.43 25.12 0.57 1.36%
spi_24_140x140 4375 26.65 50.17 40.33 16.81 -1.53 -3.66%
spi_24_180x180 4375 29.97 50.62 41.63 14.82 -0.23 -0.55%

True or Reality N° Blocks Minimum Maximum Mean SD

raw_spi_1_301[00024] 4375 14.42 67.41 41.86 81.87

The mean of “true values” is 41.86 minutes


The estimates by different spacing grid are very close to this value
No global bias for SEC estimates
SEC statistic by spacing grid
Grid spacing N° Blocks Minimum Maximum Mean SD Difference Bias
spi_24_60x60 4012 0.79 5.25 3.17 0.76 0.03 1.0%
spi_24_80x80 4221 0.83 4.97 3.18 0.69 0.04 1.3%
spi_24_100x100 4367 1.61 5.23 3.18 0.62 0.04 1.3%
spi_24_120x120 4375 1.05 4.39 3.18 0.57 0.04 1.3%
spi_24_140x140 4375 1.34 4.2 2.89 0.5 -0.25 -8.0%
spi_24_180x180 4375 1.63 4.39 3.11 0.48 -0.03 -1.0%

True or Reality N° Blocks Minimum Maximum Mean SD

SEC 4375 0.51 5.99 3.14 0.97

KWh/ton
Conclusions
• Minor biases or there are no biases, contrary as it had been assumed

• The comparisons show conditional bias

• The conditional bias is more related to the effect of information, since there is a
correction if we have more sampling density

• Therefore, the conditional bias is related to smoothing effect, a very known


property of ordinary kriging interpolator

• For the prediction of SPI™ and SEC, sampling density (and quality of the sampling)
is more relevant than the non-additivity constraints

• The results of this study allow us to conclude, that applying classical linear
averaging techniques for estimation of SPI™ is valid approach
END PRESENTATION

You might also like