Download - S2!11!10 Ignacio Velez
-
8/12/2019 S2!11!10 Ignacio Velez
1/21
Grade control sampling andSMU size optimization using
conditional simulation
-
8/12/2019 S2!11!10 Ignacio Velez
2/21
Choose a grade control sampling grid that maximizes recovery
and minimizes dilution.
Optimize the Selective Mining Unit (SMU) size to ensure mineable
envelopes can be designed with adequite recovery and minimum
dilution.
Definition of the problem
-
8/12/2019 S2!11!10 Ignacio Velez
3/21
Grade control is a fundamental part of the mining cycle.
Base of the decision Ore/Waste.
Usually it is based on what is done in similar operations.
Normaly left to inadequate or subjetive criteria.
Very little bibliography on the subject.
Starting point
-
8/12/2019 S2!11!10 Ignacio Velez
4/21
Estimation looks after local precision with the available
information.
Simulation is a technique that looks after reproducing not the
known values but the internal structure.
Simulation reproduces the sample variograms and histograms.
Not so estimation.
Estimation vs Simulation (I)
-
8/12/2019 S2!11!10 Ignacio Velez
5/21
Estimation vs Simulation (II)
Real Kriging (40 samples)
-
8/12/2019 S2!11!10 Ignacio Velez
6/21
Estimation vs Simulation (and III)
Real Simulation (40 samples)
-
8/12/2019 S2!11!10 Ignacio Velez
7/21
Based on the principles delineated in Journel and Kyriakidis
(2004)
Planning is usually done in estimated grade maps that are
normally skewed and present a softened distribution of grades.
Inadequate image of SMUs.
Traditional practice does not have the formality to analyze the
uncertainty associated to the reserve or grade control calculations.
Proposed method: Background (I)
-
8/12/2019 S2!11!10 Ignacio Velez
8/21
Method proposed will use conditional simulation to generate
possible alternatives for grade distribution.
These alternatives will be artificially sampled with GC holes and
the results used to calculate a new grade model that will be
compared with the original.
Proposed method: Background (and II)
-
8/12/2019 S2!11!10 Ignacio Velez
9/21
Proposed method: Algorithm
Analize initial data (Histogramand variogram)
Simulate n denseocurrences SGS. True
values.
Reblock to differentSMU sizes. Real
selection models
Drill with GC Holes. Futuresamples
Add errors to samples.Lower quality samples.
Calculate new blockmodel with worsened
GC holes.
Compare estimated valueswith real values and real
selection models.Results and Conclusions
-
8/12/2019 S2!11!10 Ignacio Velez
10/21
Application: Step 1
Domain 1 Acid Volcanic Rocks
Domain 2 Basic Volcanic Rocks
Domain 3 Filon Sur
Domain 4 Filon Norte
Domain 5 Oxides
Domain 6 Slates
-
8/12/2019 S2!11!10 Ignacio Velez
11/21
Application: Step 2
Samples Base de datos
Histogram Base de datos
Variogram Snowden
Parameters Snowden
50 Times
-
8/12/2019 S2!11!10 Ignacio Velez
12/21
Application: Step 3
Simulation 1 (12.5 x 12.5)
These are the real values of the mining units. Not known in paractice.
Simulation 1 (6.25 x 6.25)
Simulation 1 (4.2 x 4.2)
-
8/12/2019 S2!11!10 Ignacio Velez
13/21
Application: Step 4
9m x 9m 6m x 6m 3m x 3m
These are the real values of the GC holes. Not known in practice.
-
8/12/2019 S2!11!10 Ignacio Velez
14/21
Application: Step 5
The GC holes done over the real model do not include sampling
or assay errors.
Quality of data must be worsened.
Error model was developed with the use of historical GC data.
The model error for this case is heteroscedastic. It will depend on
the grade of the environment of the sample.
-
8/12/2019 S2!11!10 Ignacio Velez
15/21
Application: Step 6GC Holes + Errors
=
Real GC holes
Estimation
Kriging
Kriging for simulation 1 GC holes on 3x3 with errors on 6.25 x 6.25
blocks
-
8/12/2019 S2!11!10 Ignacio Velez
16/21
Application: Step 7
Simulation 1 vs
Kriging for simulation 1 GC holes on 3x3 with errors on 6.25 x 6.25 blocks
-
8/12/2019 S2!11!10 Ignacio Velez
17/21
Results (I)
50 simulations
7 grade control sampling meshes
2 error fields
3 SMU sizes (12.5m, 6.25m and 4.16m)
Total of 2,100 comparatives with real models
-
8/12/2019 S2!11!10 Ignacio Velez
18/21
Results (I)
-
8/12/2019 S2!11!10 Ignacio Velez
19/21
Results (and II)Statistical limit due to nugget effect and sampling errors to the dilution and
ore loss that can be controlled.
Dilution diminishes with sampling, ore loss does not.
SMU must be smaller than 12.5m, 6.25 m is a good compromise between size
(production rate) and reserve recovery, 4% better.
Statistical limit to the unit that can be correctly sampled. SMU below 6.25 do
not appear to have advantage. What is gained in selectivity is lost due to the
impossibility of sampling (and classifying) the blocks correctly.
Sampling mesh should be in the range of 5 x 5 m.
-
8/12/2019 S2!11!10 Ignacio Velez
20/21
Future workAnalisys of flitch mining.
Optimal bench height selection
Waste zones sampling
Underground sampling
-
8/12/2019 S2!11!10 Ignacio Velez
21/21
Thank You