Geoscience Reference
In-Depth Information
recovery. The detailed spatial distribution of these variables
permits a more holistic optimization of the mining operation.
This case study relates to BHP Billiton's Olympic Dam proj-
ect in South Australia. Two important topics are addressed
with the wealth of measurements taken at Olympic Dam:
(1) recovery and other performance variables are related to
measured rock properties through a multivariate regression
model; and (2) geostatistical models of the key rock proper-
ties are constructed by simulation.
Table 14.20
Comparison of the MIK long-term block model, MR
grade control method, and polygon-based grade control method,
March 1995-March 1996
To n s
of ore
Au
grade
Ounces
F
1
(polygonal grade control/MIK)
0.91
0.94
0.86
F
2
(plant/polygonal grade control)
1.34
0.82
1.10
F
3
= F
1
* F
2
(plant/MIK, polygonal GC)
1.22
0.77
0.95
F
1
(MR grade control/MIK)
1.13
1.00
1.13
F
2
(plant/MR grade control)
1.01
0.89
0.90
F
3
= F
1
* F
2
(plant/MIK, MR grade control)
1.14
0.89
1.02
14.7.1
Part I: Hierarchical Multivariate
Regression for Mineral Recovery
and Performance Prediction
Table
14.20
can be translated into economic gains. For
this particular operation, the MR method resulted in 34 %
more available in-situ tons, a 10 % increment of available in-
situ grade, for a net increment of 48 % in-situ metal. These
results had a major impact in the company's cash flows, oper-
ating revenues, and costs. Overall in-situ revenues increased
by US$ 11.2 million in 13 months, and the net increase in
cash flow was US$ 4.8 million, or a monthly average of
US$ 370,000, without including gains in diminished strip-
ping cost. Recall that the pit shape was unchanged. Overall
gold production jumped from approximately 65,000 ounces
annually to about 80,000 ounces. As a reference, during the
13-month period, gold prices fluctuated between US$ 370
and US$ 415 per ounce.
Aided by an improved knowledge of the geology at San
Cristóbal, the shortcomings of previously used methods for
grade estimation were better understood. The implementa-
tion of the MR method resulted in production records and
significant economic improvements to the operation. Not
only higher recovery of ore and a better ore/waste selection
overall was achieved, but there were also other operational
improvements, allowing for a reduction of the unplanned di-
lution. The MR grade control method was implemented such
that technicians with little geostatistical background can
operate and control the system. The method was in use at
San Cristóbal from February 1995 until the operation closed
down in the late 1990s.
Mineral recovery and expected plant performance are dif-
ficult to predict because they are influenced by a large num-
ber of variables such as mineralogy, grade, grain size, plant
operation parameters, etc. Often constant recovery factors
and plant efficiencies are assumed for a given mine based
on past experience and empirical rules. Such methods are
acceptable during the feasibility stages of mineral explora-
tion; however, when results of pilot plant trials are available,
statistical methods can be utilized to better predict recovery
and plant performance. In this case study, 841 bulk samples
from flotation and leach tests are used for the calibration of
a predictive model. The result is a model that can be used to
predict recovery and plant performance based on available
geometallurgical data.
Over 200 variables are available to develop a regression
model. A danger with this many variables is that the relation-
ship to recovery and plant performance variables would be
over fit. Steps must be taken to avoid over fitting. Redundant
and unimportant variables are identified and removed from
the modeling process, reducing the number of variables to
112. Through a sequence of hierarchical variable amalgama-
tion steps the variables are condensed into four major sub-
categories. A linear model based on these four amalgamated
variables provides a predictive model that is used to estimate
potential mineral recovery and plant performance over the
entire deposit. Minerals of interest in this mine include cop-
per, uranium, gold, and silver.
Plant performance is dependent on a large number of
variables, such as (1) plant feed (2) operational parameters
(3) equipment efficiencies (4) and equipment repairs. The
purpose of this case study is to relate available geometal-
lurgical data to plant performance. This is done by corre-
lating the available data to pilot plant trials. A total of 841
pilot runs are available with associated plant feed mineral-
ogy, head assays and mineral association data; the data is
described in Table
14.21
. Important plant performance in-
dicators include recovery of Cu and U
3
O
8
, acid consump-
tion (used in the leaching process), net recovery, drop weight
index (DWi) and bond mill work index (BMWi); using the
14.7
Geometallurgical Modeling at Olympic
DAM, South Australia
4
Conventional resource estimation is focused on one or a few
metals that will be sold at a profit. Increasingly, however,
it is becoming important to understand many other charac-
teristics of the ore that affect processing performance and
4
BHP Billiton—Uranium is gratefully acknowledged for allowing the
publication of this case study. This case study is based on the paper
“Geometallurgical Modeling at Olympic Dam”, by Boisvert, J., Rossi,
M. Ehrig, K., and Deutsch, C., accepted for publication in Mathemati-
cal Geosciences, 2013.
