Download as pptx, pdf, or txt
Download as pptx, pdf, or txt
You are on page 1of 18

Optimal estimation grid size to improve

profits in grade control models


Fernando Camana
Motivation

• The main objective of the grade control process is to


ensure a constant feeding of adequate material to
the plant/pad, optimizing profits, and minimizing
misclassifications.

• The grade control model is a key resource of this


process; thus, it is necessary to improve it regularly,
increasing its detail to minimize dilution, and improve
reconciliation.

• The objective of this research is to study how the


variation of the estimation grid size affects the profit,
and which is the optimum size to obtain the best
profit.

2
Introduction

• Profits depend on grades, metal prices, mining costs,


etc., so how to improve the profits in grade control
models?
– A high-resolution estimation produces higher
grades, better block selectivity and decreases the
dilution, improving the profits.

• High-detailed block models could require excessive


computational and time resources.

• The optimal size for the estimation grid can be defined


based on the most precise numerical model without
excessive use of computer resources (Vasylchuk,
2016).

Profit Ore Waste Grade


GSS
(USD) (t) (t) (g/t)
Control 898,160 661,770 16082 0.37
2.00 849,941 665,582 12270 0.37

3
Estimation grid size for grade control models

• On most occasions, the grid size is selected


according to long and middle term models (SMU)
and blast-hole data (Leuangthong, Neufeld &
Deutsch, 2003).

• For grade control models, the large block sizes do


not have the selectivity and resolution needed. On
this context is preferable to use the data spacing as
reference for the estimation grid size.

• The GSS is the ratio between the estimation grid


size and data spacing. For a GSS of 0.25, the
estimation errors are minimized within acceptable
processing time (Vasylchuk & Deutsch, 2015).

4
Grade control model resolution

• The block model resolution contributes to profits. The


lower GSS the higher model resolution, higher grades, SWST

and thereby the best profit.


SWST Ton: 8438.8
Ton: 9201.6
 
 

• A high-resolution model permits a better definition of dig


limits, thus a high selectivity and low dilution.

• At the same time, a high-resolution model could result


in unrealistic dig limits which decrease the efficiency of SWST
Ton: 8051.4
SWST
Ton: 7907.6

the mining equipment.  


 
 

• But high-resolution models implies more processing


time, and in the grade control process, time is gold.

5
Methodology

1. Data acquisition
– Gaussian simulation
– Sampling
2. Kriging
– Estimation parameters
3. Profit calculation
– Marginal cutoff
– True profits
– Ore and waste tonnage
– Time usage

6
Data acquisition – SGS simulation

• 300 distributions simulated (SGS) and 50


realizations per each distribution.

• Each realization transformed to log normal


distribution and compiled.

• Work area of 200mx150m, single bench.

• Resolution of 1mx1m.

7
Data acquisition - Sampling

• A realistic blast-hole grid.

• Average data spacing of 6 meters.

• Non-orthogonal arrangement.

• All distributions sampled using the same grid.

• 300 different datasets with 1058 samples.

8
Kriging – Estimation parameters

• Ordinary kriging (KT3D). 1


2
Parameters for KT3DN
********************
3 START OF PARAMETERS:
4 sampling.dat -file with data
5 1 2 3 4 5 0 -columns for DH, X, Y, Z, var

• GSS changed sequentially and based on multiples 6


7
-998.0
0
1.0e21 -trimming limits
-option: 0=grid, 1=cross
8 xvk.dat -file with jackknife data
and submultiples of data spacing. Includes a control 9
10
2 3 4 5 0
kt3dn_dataspacing.out
-columns for X, Y ,Z, vr
-data spacing analysis output

grid. 11
12
0
0
10.0
100 0
-number to search and comp
-debugging level: 0,3,5,10
13 kt3dn.dbg-nkt3dn.sum -file for debugging output
Kriging Grid Size GSS 14
15
OK_12m_sim1.out
17 103 12
-file for kriged output
-nx,xmn,xsiz
1x1m Control 16
17
13
1
353
10
12
1
-ny,ymn,ysiz
-nz,zmn,zsiz
1.5x1.5m 0.25 18
19
3
24
3
120
1
12 1
-x,y and z block disc
-min, max data for kriging
3x3m 0.50 20 0 0 -max per octant, max per DH
21 50.0 50.0 10.0 -maximum search radii
6x6m 1.00 22 45 0.0 0.0 -angles for search ellipsoid
23 1 -0=SK, 1=OK
12x12m 2.00 24 2.302 0.6 0.8 -mean, corr., var.

• Number of samples, search radio and block


discretization, the same for all datasets.

• Variogram model specific for each dataset.

9
Kriging - Results

• Higher GSS produce low-resolution models, but for


GSS 0.25 and the control grid, this improves
notoriously.
• Although the sampling was quite representative, the
kriging cannot replicate exactly the "true" distribution.
• The processing time is what we expected, the lower
the GSS, the higher time.

GSS Time (s)


Control 15.11
0.25 6.91
0.50 1.93
1.00 0.57
2.00 0.16

10
Marginal cutoff

• The marginal cutoff define the ore and waste blocks.

Profit parameters
Mining Cost 3.4 $/t
Processing Cost 9.25 $/t
Recovery 80%
Au Price 46.1 $/g
Density 2.13 g/cm3
• The ore and waste blocks increase as the GSS
decreases, but they still far from the true distribution.

11
Profit calculation

• We must calculate the profit respect to the true


distribution.

• To do that, all block model must be sub-blocked to


coincide with the true distribution.

• The profit is calculated using the previous ore limits


and the grades from the true distribution.

12
Profit calculation - Results

• All results must be standardized by dividing them by the control grid results.

• The Control grid has the highest profit, which decreases as the GSS increases.

GSS Profit Ore Waste


Control -
0.25 -0.12% 0.04% -0.96%
0.50 -0.33% 0.25% -6.37%
1.00 -1.13% 0.94% -24.02%
2.00 -3.09% 2.37% -57.84%

13
Profit calculation - Results

• There is a big difference respect true distribution, specially for waste. This could due to the smoothing of
kriging, which reduces the extreme values and increments the number of average grade blocks.

GSS Profit Ore Waste


True - - -
Control -7.50% 8.14% -64.26%
0.25 -7.60% 8.18% -64.58%
0.50 -7.79% 8.41% -66.36%
1.00 -8.47% 9.16% -72.18%
2.00 -10.17% 10.70% -83.83%

14
Conclusions

• GSS 0.25 and control grid have practically the same


profit results and similar block model resolution,
being the main difference, the processing time.

• We conclude that GSS 0.25 is the optimal estimation


grid, producing the highest profit within an
acceptable time usage.
Profit Ore Waste
GSS
(USD) (t) (t)
True 1,063,442 607,583 70,269
Control 1,014,011 652,398 25,454
0.25 1,013,638 652,441 25,411
0.50 1,011,802 652,547 25,304
1.00 1,005,800 655,614 22,237
2.00 982,626 665,582 12,269

GSS Profit Ore Waste Time


Control - - - 1.00
0.25 -0.12% 0.04% -0.96% 0.46
0.50 -0.33% 0.25% -6.37% 0.13
1.00 -1.13% 0.94% -24.02% 0.04
2.00 -3.09% 2.37% -57.84% 0.01

15
Conclusions

• The profit difference between the highest and the


lowest GSS is on average 2.5%, but in some cases
can reach 17%.

• This seems low but corresponds to a small area


(200m x 150m).If this area is extrapolated to a whole
bench (2kmx1km), the improve in profits can reach
almost USD 1.8M.

GSS Difference (%) Profit (USD) Profit (USD) Diff.


(200mx150m) (2kmx1km) (USD)
True - 1,000,000 66,670,000 -
Control -7.50% 925,000 61,669,750 -
0.25 -7.60% 924,000 61,603,080 -66,670
0.50 -7.79% 922,100 61,476,407 -193,343
1.00 -8.47% 915,300 61,023,051 -646,699
2.00 -10.17% 898,300 59,889,661 -1,780,089

16
Conclusions

• The difference in the kriging results in respect to the


true distribution is due to the lack of information and
the limitations of the estimation method. Thus, it is
necessary to improve the estimation or implement a
better method, like simulation.

• The higher GSS the greater ore tonnage, but this


does not mean a better profit. The smoothness of
kriging affects more the higher GSS, decreasing high
grades and with that the profits.

• The processing time must be considered as a


reference because they correspond to laboratory
conditions.

17
18

You might also like