You are on page 1of 10

Simulation of Autonomous Mine Haulage Trucks

Juliana Parreira and John Meech


Norman B. Keevil Institute of Mining Engineering,
The University of British Columbia, Vancouver, BC, Canada, V6T1Z4
Corresponding author: (jameech@dccnet.com)

Abstract
Driverless haulage trucks have recently been developed for open pit mines. To predict the
benefits of an Autonomous Haulage Truck (AHT) system, a deterministic/stochastic model has
been created to compare an AHT system to a manual system by determining benchmarked Key
Performance Indicators (KPIs) such as productivity, safety, breakdown frequencies, maintenance
and labor costs, fuel consumption, tire wear, and cycle times. The goal of this paper is to
describe the model developed to describe vehicle motion and fuel consumption sub-models that
function within a virtual 24/7 open pit mine operating with 9 trucks and 2 shovels to move ore to
a crusher and waste rock to a dump.
The model extends conventional shovel/truck simulation into a variety of truck sub-systems such
as truck movement, driver behaviour, fuel consumption, and tire wear to capture the mechanical
complexities and physical interactions of these sub-systems with the mine environment on a 24/7
time basis. Running the model in identical scenarios for the two cases allows comparison of
benchmarked KPIs that demonstrate the utilization and adaptability of an AHT.

Introduction
Over the past decade, mining companies have been challenged by increasing global demand,
fluctuating commodity prices, difficult-to-access minerals due to location and harsh
environments, as well as social and environmental issues. Automation is one tool that can be
used to maximize production, reduce costs, and create a safer work environment in a customerdriven economy that demands rapid response. In particular, automation of open pit haulage
systems is receiving attention to remove workers from potential dangers, to reduce costs, and to
help improve mining equipment efficiency.
This research project is examining how an open pit mine can adapt itself to an autonomous
haulage fleet by creating a tool to compare conventional and Autonomous Haulage Truck (AHT)
systems. What level of improvement in KPIs can be achieved using an AHT? What aspects of a
mine haulage system must change to adapt to an AHT? In order to answer these questions, two
discrete-event models (one manual and one autonomous) were developed to predict
benchmarked KPIs: productivity; safety; maintenance and labour costs; cycle times; fuel
consumption; and tire wear under different road and load conditions. The scope of the work is
limited to offline simulation software that project managers can use to guide decision-making on
the possible application of an AHT system in a specific mine.

Model Overview
The model was built using the ExtendSim simulation software suite, a graphical discrete-event
tool that can breakdown a network into unique components each having specific time delays and
characteristics with respect to maintenance, speed, fuel consumption, braking, acceleration, etc.
A system database is used to input model parameters such as route characteristics, weather,
equipment fleet parameters, driver behaviour, and stochastic sets (averages and variances) that
characterize different interruptions and delays. The data is held in a RAM-resident database that
opens, saves, or closes whenever the model is opened, saved, or closed respectively.
The test model consists of two shovels, one digging waste and one assigned to ore production.
According to a specified stripping ratio, one third of the fleet (9 trucks in total) are set to work
with the waste shovel with the other six are assigned to the ore shovel. The model can designate
dedicated service or trucks can be reassigned on-the-fly after dumping depending on the current
delay situation. All trucks in the model are CAT 793D, each having a nominal GMW of 383,749
kg, a total net power of 1,743 kW, and equipped with standard radial tires 40.00R57. Different
trucks can be easily configured through the database.
For each segment of a haulage route, the movement of each truck, its fuel consumption, and tire
wear are determined using a continuous (or deterministic) approach. The objective is to handle
speed and acceleration control issues at any point in time and to calculate fuel consumption and
tire wear for each time step (t = 0.1 seconds). The Gross Machine Weight (GMW), truck
resistance forces, traction, and drive axle forces are used to calculate the resultant force causing
movement. For each t, a new force and acceleration is calculated until the cumulative travelling
distance equals the total segment distance. At the end of each road segment, data is initialized for
the next segment with simulation continuing until the truck reaches its final destination.
The model is a partial haulage system at a real mine. A total of 19 road segments are distributed
over 4 haulage routes: Ore Shovel to Crusher; Waste Shovel to Dump; Dump to Parking; and
Crusher to Parking. If the truck is set at the beginning of the simulation to attend the waste
shovel, its overall route is Waste Shovel to Dump, while for trucks set to work on ore production,
the route is Ore Shovel to Crusher. When a truck arrives at a shovel queue, it waits for the shovel
to become free. Values for spotting and loading times are set stochastically. After loading, the
truck leaves the shovel and travels to its dump site. At the dump or crusher queues after waiting
for the site to become free, the truck reverses and dumps with times selected stochastically.
Queuing times depend on the presence of other trucks in the queue. After unloading, the truck
returns to its original route and continues in this loop until an event or stochastic delay occurs or
the shovel is down for maintenance. In that case, the truck is reassigned to maintenance or the
other shovel. The model runs for a simulated time of 14-days of continuous operation.
A refuelling delay takes place whenever the tank level falls below a set minimum value (10% of
a full tank) when the truck is at a dumping location. When that occurs, the truck dumps its
payload and is reassigned to the refuelling station (Parking). A stochastic refuelling time is
selected to characterize the time to complete this event. If a truck requires maintenance or a
driver needs a break (lunch, coffee, or shift change), the truck drives to parking, but only after its
load is dumped. When running in Autonomous Haulage mode, driver breaks do not happen, but
other event delays may occur such as tire cool-down.

After a truck is repaired, values for MTBF (mean time to failure) and MTTR (mean time to
repair) are reset. During the simulation, data is stored and managed in the internal database and
the relevant run results are exported to an Excel template spreadsheet when the simulation run is
completed. The model flowchart is shown in Fig. 1.

Figure 1: Overview of Model Flowchart.

Truck Movement
The truck movement model is based on the assisting and opposing forces such as Rimpull
available, usable, required (FRR) and drag forces (FD). Rimpull available is the force produced by
the engine delivered at the point that the tires contact the ground. Usable Rimpull deals only with
the weight on the drive wheels and the friction between the tire and the ground [King Fahd
University, 2004/2005]. Feffective is the lower value between rimpull available and rimpull usable.
Knowing the resultant forces and payload, acceleration can then be calculated [Parreira, et.al,
2011]. Acceleration is calculated deterministically for each time step and distinguishes between
acceleration responsible for movement and that responsible for fuel consumption. When a truck
is moving on a flat road, acceleration responsible for truck movement is:
Accresultant = (Feffective - Fresist) / M

Eq. 1.

where Feffective is the assisting force propelling the truck; Fresist is the force opposing movement;
and M is the truck weight and payload. If the grade is positive, Fresist = Fgravitational + Frolling Fdrag.
If the grade is flat, the equation becomes Fresist = Frolling Fdrag. The value of Fdrag depends on wind
direction and speed which may either oppose or assist with truck movement.

Figure 2: Truck Forces when Moving on a Grade = 0%.

Figure 3: Truck Forces when Moving on a Grade < 0% and > 0% respectively.
If the grade is negative and the gravitational force (normal component of truck weight plus
payload) is sufficient to trigger truck motion, then this force is used as the truck assisting force.
However, if the weight is not enough, an extra force (Rimpull available) must be produced by the
engine. After calculating Accresultant, the model applies a factor to this value dependent on driver
behaviour. The value is limited to the maximum acceleration of the type of driver in question as
discussed below. When a truck reaches the velocity set point for the road segment or the limiting
velocity based on driver behaviour, Accresultant is set to a value close to zero. A slight variation
exists, also dependent on driver behaviour.

Driver Behaviour
The objective of the driver model is to generate stochastic differences in driver behaviour to
obtain output ranges for fuel consumption, tire wear, cycle times, and production for a manuallyoperated system that mimics real mine site data. These ranges can then be compared to that
achieved by a simulated AHT system in which the variances and deviations from normal or
accepted results are significantly reduced.
A parameter called the Aggressiveness Factor is used to represent how a person drives. A driver
can be aggressive, normal, or passive. A driver is very stable, stable, or variable such that 9
different behaviours result. This factor is set at the beginning of the simulation with the fleet
crew being distributed amongst these 9 types. Autonomous trucks are set to the normal-stable
driver behaviour with a smaller deviation than a human. The Aggressiveness Factor is applied to
the acceleration, velocity, and reaction time set points for each driver/truck which can trend up or
down during the shift [Meech et al., 2011] to reflect slight changes in behaviour. Some drivers
become more aggressive while others become more passive. Fig. 4 shows 3 different driver
behaviours over a simulated 1,000 m segment with a -12% effective grade. There are 9 different
driver types but 3 are dominant: passive-very stable; normal-stable; and aggressive-variable.

Figure 4. Acceleration and velocity of different drivers over a 1,000 m haulage segment
with a 12% effective grade from ExtendSim Model.

Fuel Consumption
For this sub-model, truck manufacturer data combined with fundamental physics is used to
estimate fuel use in L/hour. Instantaneous velocity is calculated for each time interval t. By
knowing the desired velocity, engine speed is calculated and then correlated to Power and BSCF
[Parreira, et.al, 2011]. Fuel consumption is estimated by the following equation:
Cf = Csf (Powerused / f)

Eq. 2.

where:
Cf
= fuel consumption rate (L/h)
Csf
= Brake Specific Fuel Consumption (BSFC) for time interval t (g/kWh) (see Fig. 5)
f
= fuel density (g/L)
Powerused = power produced by the engine during time interval t (kW) based on Fig. 5.

Figure 5. Instantaneous Power and BSCF vs. Engine Speed - from fuel consumption model.
In the example shown in Fig. 6, using gear 1 at 1739 rpm consumes 451 L/h. However, if the
driver switches to gear 2, fuel consumption decreases to 383 L/h (a reduction of 15%). The
model gives fuel consumption according to which gear is in use. When human behavior is
eliminated from the system, there is an opportunity to redesign haulage roads based on this
increased efficiency and more precise engine speed changes and acceleration set points.

Figure 6. Using the correct gear leads to a 68 L/h reduction in fuel consumption.

Tire Wear
In general, tire wear for a 40.00R57 radial tire on a CAT 793D truck is about 75 mm over an
operating life of 6,000 to 6,500 hours. Tire treads are measured on a regular basis during routine

maintenance and tires are removed from service when the depth falls below about 25 mm from
an initial depth of 100 mm.
We have created a sub-model based on fuzzy logic that correlates tire wear to truck velocity and
payload. The model can be calibrated to reflect known operating conditions at any specific mine
site and can be adjusted throughout the life of a set of tires to reflect the current tire condition on
a truck. The model also predicts changes in tire temperature as a truck operates in order to
account for wear rate differences when hauling under full load or returning to the shovel at a
higher velocity with zero load. The idling time during loading, dumping, and queuing is also
accounted for in the temperature model. Fig. 7 shows the contour plot of tire wear rate in
mm/10,000km (as a % of maximum), depending on payload and velocity.

Figure 7. Tire Wear vs. Payload at Different Velocities Accumulation Defuzzification method.
Calibration Factors: Maximum Speed = 40 kph; Maximum Payload = 400 t;
Maximum Tire Wear = 10 mm / 10,000 km
Note that the two boxes that represent typical ranges for full-load and empty conditions indicate
similar tire wear rates since one is at a higher GMW while the other is at a higher velocity. Three
parameters are required to calibrate the model - maximum tire wear rate in mm/10,000 km,
maximum velocity in kph, and maximum payload in tonnes.
Many mines use tire suppliers' TKPH model to reduce tire failures and avoid overheating tires
during operation. Using this model provides a reliable and practical constraint beyond which a
truck must stop to allow tires to cool-down or the truck is reassigned to a longer haulage route
and restricted to 4th gear (to prevent speeding). When the real-time TKPH calculation returns to
normal, the truck is reassigned to normal service. While this may help reduce failures from cut or
blown tires, it doesn't provide a real-time measure of temperature or wear rate. Furthermore,
restricting gear changes will lead to increased fuel use.
We believe there are significant advantages to gain in monitoring tire temperature directly and
then used with an empirical model to provide a real-time measure of tire wear for 40.00R57
radial tires on a CAT 973D truck. Fig. 8a and Fig. 8b show the theoretical model predictions of
how a tire heats up and cools down under continuous cycle conditions. The model is designed so
it takes about 60 minutes of constant driving at 16 kph under target payload conditions to arrive
at a steady state temperature of 75C from an initial ambient temperature of 35C, while cooldown under idling conditions requires about 120 minutes to return to ambient conditions.

(a)

(b)

(c)
Figure 8. Effect of idling time on tire temperature during the haulage cycle. Idling time
as % of total cycle time: (a) 14.7%; (b) 9.3%; (c) 9.3% at elevated velocities.
Fig. 8a shows that when idling time represents about 15% of total cycle time, a steady state
temperature of 61 C is reached, whereas when the idling time is only about 10% of total cycle
time, the steady state temperature rises to 80 C as in Fig. 8b. Fig. 8c shows what can happen
when truck velocities are increased to 19 and 38 kph for loaded and empty conditions
respectively. Although the rise in temperature initially is lower than in Fig. 8b, the system is
unstable with the temperature continuing to rise well above the danger point of 90 C after 9
cycles. Clearly, this truck will require a cool-down period.

Results and Validation


The model output shown in Tables 1 and 2 is based on a 14-day work period; the crew make-up
was set to 33% aggressive, 44% normal, and 22% passive. In the autonomous model, all trucks
were 100% normal, i.e. best drivers, with a lower variance. The AHT system shows a 14.4%
improvement in production, a decline of 12.9% in fuel consumption (L/t) and a 7.2% decline in
tire wear (mm/t). The tire wear model is currently being validated against real mine data. Fuel
consumption and truck movement are both close to measured data.
The set points for acceleration and speed according to driver behaviour have also been calibrated
against 12 drivers at the mine; those who drive slower and those who are more aggressive. The
tire wear model is the most difficult one to validate since most mines simply measure tire wear
based on the TKPH approach. The tire manufacturer sets a limiting TKPH value and the mine
dispatch system verifies if a truck TKPH is higher than this set point based on haulage distances
and payloads. We believe improvements can be derived from direct measurement of tire
temperature to predict tire wear rates. An AHT fleet will work 24/7, so sensors on belted tires are
required to ensure tire cool-down is optimized as an AHT fleet will operate for longer durations.
Table 1: Comparison of different drivers - from model template spreadsheet.
Element
Total Cycle Time

Passive
53.5
5.6
10.5

Normal
47.3
5.3
11.2

Aggressive
42.5
4.8
11.3

Autonomous
50.5
5.1
10.1

Fuel Consumption (Idling) L/hr


Fuel Consumption (Full) L/hr
Fuel Consumption (Empty) L/hr

26
366
134

27
415
138

28
437
179

27
410
127

Total Litres/cycle
% difference

261.2
-9.8

289.5
0.0

314.4
8.6

286.5
-1.0

Idling and
Spotting Time

min.
min.
%

Total Litres/tonne
% difference

<------------------------------ 1.55 ------------------------------>


<-------------------------------- 0.0 ----------------------------->

1.35
-12.9

Tire Wear Rate (Idling) mm/hr


Tire Wear Rate (Full) mm/hr
Tire Wear Rate (Empty) mm/hr

0.0032
0.0177
0.0096

0.0032
0.0253
0.0120

0.0032
0.0299
0.0164

0.0032
0.0212
0.0106

Tire Wear (mm/hr)


% difference

0.0139
-6.7

0.0149
0.0

0.0152
2.3

0.0138
-7.2

Tire Wear (mm/ 10,000 tonnes)


% difference

<------------------------------ 0.570 ---------------------------->


<-------------------------------- 0.0 ----------------------------->

0.536
-6.0

It is interesting to note in Table 1 that passive drivers have lower tire wear than do the aggressive
drivers and they use less fuel. Note that the AHT system matches the tire wear of the passive
drivers and uses less fuel than the average driver. This translates to an 8.1% reduction in fuel
used per tonne of material moved. Assumptions about maintenance have held delay times due to
repairs relatively close to those measured for the manual system except for unplanned
maintenance which is expected to decline by about 10% as can be seen in Table 2.

Table 2: Productivity of Manual Haulage versus AHT System - model template spreadsheet.
Manual
Autonomous
%Difference
hours
Hours
Number of Cycles/day
19.82
22.35
14.00
Ave. Total Cycle Time/truck - min.
46.8
50.5
5.71
Total Haulage Time/day/truck
15.46
18.79
23.13
Shift Change/day/truck
0.60
0.00
-100.00
Coffee and Break Time/day/truck
2.50
0.00
-100.00
Average Process Delay Time/day/truck
1.64
1.60
-2.44
Unplanned Maintenance/day/truck
1.16
1.04
-10.34
Planned Maintenance/day/truck
2.64
2.67
1.14
Percent Utilization (%)
63.6
78.3
23.14
Total Production (tonnes) /day/truck
4,235
4,844
14.35
Ave. Production / cycle/truck
213.7
216.7
1.40*
* this difference should be close to zero, meaning the simulation needs to be run for more than 14 days.
Element

Conclusion
This model is in verification/validation process and when finished, it will be able to provide a
precise comparison between manual and autonomous haulage trucks regarding KPIs such as
productivity, safety, maintenance, labour costs, cycle time, fuel consumption, and tire wear. The
model is structured in a way to allow quick adjustment to simulate any open pit mine that is
considering an autonomous haulage fleet in the future.

Acknowledgment
The authors wish to thank BHP-Billiton Nickel West Division for supporting this research.

References
King Fahd University of Petroleum & Minerals, Construction Engineering & Management
Department: Construction Equipment & Methods (2004). Course: CEM 530 (Spring
2004/2005)
Meech, J., Parreira, J. (2011) . An interactive simulation model of human drivers to study
autonomous haulage trucks. Complex Adaptive System Conference. Procedia Computer
Science, 6, 118-123.
Parreira, J., Meech, J. (2011) . Autonomous Haulage Systems Justification and Opportunity.
Autonomous and Intelligent Systems, International Conference IS-2011. Springer, 63-72

You might also like