Including stockpiles into

mathematical programming models

for mine planning
Felipe Ferreira, Ms. Sc. Student, Universidad Adolfo Ibañez, Santiago, Chile
Eduardo Moreno, Associate Professor, Universidad Adolfo Ibañez, Santiago,
Chile.
 Introduction
• Mine Planning software. What do they optimize?
• Introduction of optimization models in the 60’s
– Ultimate Pit Limit, Lane, Lerchs-Grossman’s 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 Whittle’s 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 can’t be used in great size (let’s say real size) instances

• We present three linear models for including stockpiles in

long-term mine planning
 Non-linear model: Blocks
• For each block 𝑏 ∈ 𝐵
– 𝑤𝑏 : total tonnage 𝑚𝑏
Metal Grade:
– 𝑚𝑏 : metal tonnage 𝑤𝑏
• Time period 𝑡 ∈ 𝑇
𝑒
• If block 𝑏 is extracted in 𝑡: 𝑥𝑏,𝑡 ∈ {0,1}
𝑝
– Fraction of block 𝑏 sent to processing plant: 𝑥𝑏,𝑡
𝑠
– Fraction of block 𝑏 sent to stockpile: 𝑥𝑏,𝑡
• This lead us to some constraints:
𝑝 𝑠 𝑒
– Block destination: 𝑥𝑏,𝑡 + 𝑥𝑏,𝑡 ≤ 𝑥𝑏,𝑡
– Block must be extracted in one period only:
𝑒
𝑥𝑏,𝑡 ≤1
𝑡∈𝑇
 Non-linear model: Stockpile
• Assumption: Extracted Ore arrives to stockpile at the end of period 𝑡, and
Ore is reclaimed from stockpile at the beginning of period 𝑡
• Variables:
– Ore, metal available in stock at the end of period 𝑡: ost , 𝑎𝑡𝑠
p 𝑝
– Ore, metal sent to mill from stock at the beginning of period 𝑡: ot , 𝑎𝑡
• Then:
𝑝 𝑠
– 𝑜𝑡 ≤ 𝑜𝑡−1
𝑝 𝑠
– 𝑎𝑡 ≤ 𝑎𝑡−1
t t+1
• Finally the amount of Ore and Metal in stockpile a the end of period 𝑡 is:
𝑠 𝑠
𝑤𝑏 ∗ 𝑥𝑏,0 𝑡=0 𝑚𝑏 ∗ 𝑥𝑏,0 𝑡=0
𝑏∈𝐵𝑡 𝑏∈𝐵𝑡
𝑜𝑡𝑠 = 𝑎𝑡𝑠 =
𝑠 𝑝 𝑠 𝑠 𝑝 𝑠
𝑜𝑡−1 − 𝑜𝑡 + 𝑤𝑏 ∗ 𝑥𝑏,𝑡 𝑡>0 𝑎𝑡−1 − 𝑎𝑡 + 𝑚𝑏 ∗ 𝑥𝑏,𝑡 𝑡>0
𝑏∈𝐵𝑡 𝑏∈𝐵𝑡
 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:
𝑝 𝑠
𝑎𝑡 𝑎𝑡−1
𝑝 ≤ 𝑠 ∀𝑡 ∈ 𝑇
𝑜𝑡 𝑜𝑡−1

Average Metal Grade

𝑚1 𝑚2 𝑚3 𝑚4 𝑚5
𝑤1 𝑤2 𝑤3 𝑤4 𝑤5

Stockpile
 Non-linear model: Objective function and other
Constraints
Incomes Expenses

Precedence constraint

Extraction Capacity

Processing Capacity
 Linear Models: Upper Bound
• Blocks are stockpiled (and reclaimed) independently
from each other
• There’s no instant mixing
• Infeasible solution!
 Linear Models: Lower Bound
• Material reclaimed from stockpile has a fixed metal grade 𝐿
𝑝 𝑝
– We replace the instant mixing constraint with: 𝑎𝑡 = 𝐿 ∗ 𝑜𝑡
L-Bound model
• Blocks sent to stockpile must have a metal grade above 𝐿

s mb
xb,t =0 ∀b ∈ B t. q. : <L
wb

L-Average model
• The cumulative average metal grade of the blocks sent to the
stockpile must be at least 𝐿
s s
𝑚𝑏 ∗ xb,𝑡 ′ ≥ 𝐿 ∗ 𝑤𝑏 ∗ xb,𝑡 ′

𝑡 ′ ≤𝑡 𝑏∈𝐵 𝑡 ′ ≤𝑡 𝑏∈𝐵
 Results
• Instance 1: Marvin (MineLib)
– Solved with a fixed extraction sequence (so we can solve the non-
linear 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 Material sent to mill

Non-linear

L-Average

L-Bound

Mill Stockpile Waste dump

 Results
• Instance 2: Tampakan
– Great Size Instance, we couldn’t 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 NPV % difference

Upper Bound $ 4,848,040,000 -
L-Average $ 4,677,720,000 -3.51%
L-Bound $ 4,451,700,000 -8.18%
No Stockpile $ 4,296,550,000 -11.38%
 Solution Analysis
Incoming material to mill, and arsenic grade in it (L-Average Solution)
80.000.000 160
70.000.000 140
60.000.000 120
Tonns of Material

Arsenic Grade
50.000.000 100
40.000.000 80
30.000.000 60
20.000.000 40
10.000.000 20
- -
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

L-Average
4000
Destination of blocks 3500
extracted on first 3000
Arsenic Grade

2500
period 2000 Planta
Mill
1500 Stockpile
1000
500 Desecho
Waste
Cut-off grade: 0.21% 0 Dump
0 0,5 1 1,5 2 2,5 3
Metal Grade
 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
$ 877,732,900 +3.62%
(no stock)
Optimal schedule
$ 911,356,530 +7.59%
(with L-average stockpile)
 Solution Analysis
70.000.000

60.000.000
Destination
Tonns of Material

50.000.000
of Extracted
Material 40.000.000 Waste Dump

30.000.000 Sento to Stock

Sent to Mill
20.000.000

10.000.000

-
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Period

70.000.000 1,6

60.000.000 1,4
1,2
50.000.000
Tonns of Material

Material

Metal Grade
1
sent to 40.000.000
0,8
Stock to Mill

mill, and 30.000.000

0,6
Sent to Mill

metal Metal Grade in Mill

20.000.000
0,4
grade in it 10.000.000 0,2
- 0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Period
 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