Sei sulla pagina 1di 17

Including stockpiles into

mathematical programming models


for mine planning
Felipe Ferreira, Ms. Sc. Student, Universidad Adolfo Ibaez, Santiago, Chile
Eduardo Moreno, Associate Professor, Universidad Adolfo Ibaez, Santiago,
Chile.

Introduction
Mine Planning software. What do they optimize?
Introduction of optimization models in the 60s
Ultimate Pit Limit, Lane, Lerchs-Grossmans Algorithm
Johnson 1968
More details: Newman et al. (2010) and Osanloo et al.
(2008)

Many models and methods reach better solutions


than commercial software.

Literature Review
Stockpile has great importance in mining operations
Few authors include the stockpile option in long-term optimization models
Bley et al. (2012a)
Difficulty in modelling stockpiles: mixing behavior

Ramazan & Drimitrakopoulos (2013)


Stochastic method with stockpile option

Smith & Wicks (2014)


Use an optimization model with stockpile option for medium-term planning.

Geovia Whittles manual


Stockpile withdrawals are considered to be at the average grade of material
sent to it

Literature Review
Tabesh et al. (2015)
Shows a non-linear model with stockpile option
Linear model: uses a lot of stockpiles with predefined metal grades

Bley et al. (2012b)


Introduces two non-linear models
They consider instant mixing of the material send to the stockpile
Non-linear and non-convex restriction
Models cant be used in great size (lets say real size) instances

We present three linear models for including stockpiles in


long-term mine planning

Non-linear model: Blocks


For each block
: total tonnage
: metal tonnage

Metal Grade:

Time period

If block is extracted in : ,
{0,1}

Fraction of block sent to processing plant: ,

Fraction of block sent to stockpile: ,

This lead us to some constraints:

Block destination: , + ,
,

Block must be extracted in one period only:

,
1

Non-linear model: Stockpile


Assumption: Extracted Ore arrives to stockpile at the end of period , and
Ore is reclaimed from stockpile at the beginning of period
Variables:
Ore, metal available in stock at the end of period : ost ,
p

Ore, metal sent to mill from stock at the beginning of period : ot ,


Then:

t
t+1
Finally the amount of Ore and Metal in stockpile a the end of period is:

,0

,0

=0

1
+

=
>0

=0

1
+

>0

Non-linear model: Instant Mixing


Other assumptions (but important):
Blocks sent to stockpile are instantly mixed reaching
homogeneity
Other processes are not considered (for example:
Comminution Process)

Instant mixing constraint:

1

1

Average Metal Grade

1 2 3 4 5
1 2 3 4 5

Stockpile

Non-linear model: Objective function and other


Constraints
Incomes

Precedence constraint

Extraction Capacity
Processing Capacity

Expenses

Linear Models: Upper Bound


Blocks are stockpiled (and reclaimed) independently
from each other
Theres no instant mixing
Infeasible solution!

Linear Models: Lower Bound


Material reclaimed from stockpile has a fixed metal grade

We replace the instant mixing constraint with: =

L-Bound model
Blocks sent to stockpile must have a metal grade above
s
xb,t

=0

b B t. q. :

mb
<L
wb

L-Average model
The cumulative average metal grade of the blocks sent to the
stockpile must be at least
s
xb,

s
xb,

Results
Instance 1: Marvin (MineLib)
Solved with a fixed extraction sequence (so we can solve the nonlinear model with SCIP)

Variation of processing capacities to observe economical


impact of stockpiles
Cap. UB
Non-Linear
L-Average L-Bound No stock.
60%
2.1% $ 742,292,000
-0.3%
-4.8%
-11.8%
70%
1.3% $ 820,693,000
-0.1%
-3.8%
-8.2%
80%
0.6% $ 882,863,000
0.0%
-2.5%
-5.1%
90%
0.3% $ 928,833,000
0.0%
-1.4%
-2.9%
100%
0.1% $ 961,253,000
0.0%
-0.7%
-1.3%

Solution Analysis

Extracted material destination


Non-linear

L-Average

L-Bound

Mill

Stockpile

Waste dump

Material sent to mill

Results
Instance 2: Tampakan
Great Size Instance, we couldnt use the non-linear model
High presence of arsenic contaminant capacity constraint for this
element
Stockpile used for lowering arsenic average level in material sent to
mill

Model
Upper Bound
L-Average
L-Bound
No Stockpile

NPV
$ 4,848,040,000
$ 4,677,720,000
$ 4,451,700,000
$ 4,296,550,000

% difference
-3.51%
-8.18%
-11.38%

Solution Analysis
80.000.000

160

70.000.000

140

60.000.000

120

50.000.000

100

40.000.000

80

30.000.000

60

20.000.000

40

10.000.000

20

Arsenic Grade

Tonns of Material

Incoming material to mill, and arsenic grade in it (L-Average Solution)

0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50
Period

Destination of blocks
extracted on first
period
Cut-off grade: 0.21%

Arsenic Grade

L-Average
4000
3500
3000
2500
2000
1500
1000
500
0

Planta
Mill
Stockpile
Desecho
Waste

Dump
0

0,5

1,5
Metal Grade

2,5

Considering extraction decision


Instance: Marvin (same than before)
Our model reaches a higher NPV, even without using a
stockpile, than commercial software Whittle.
Solution

NPV

Variation

Whittle

$ 847,035,400

Whittle + L-average

$ 855,442,430

+0.99%

Optimal schedule
(no stock)

$ 877,732,900

+3.62%

Optimal schedule
(with L-average stockpile)

$ 911,356,530

+7.59%

Solution Analysis
70.000.000
60.000.000

Tonns of Material

Destination
of Extracted
Material

50.000.000
40.000.000

Waste Dump

30.000.000

Sento to Stock
Sent to Mill

20.000.000
10.000.000
1

10 11
Period

12

13

14

15

16

17

18

70.000.000

1,6

60.000.000

1,4

19

20

1,2

50.000.000

40.000.000

0,8
30.000.000

0,6

20.000.000

0,4

10.000.000

0,2

0
0

8
9
Period

10

11

12

13

14

15

16

17

Metal Grade

Material
sent to
mill, and
metal
grade in it

Tonns of Material

Stock to Mill
Sent to Mill

Metal Grade in Mill

Conclusions
Stockpile use increases NPV of mine operations
Linear model with stockpile option
Practical way, can be used in large instances
Behaves similar than non-linear models

Optimization models defy classical ways to perform


mine planning
Stockpile use affects extraction sequence and block
destination, but this is not considered by Com. Softwares

Potrebbero piacerti anche