You are on page 1of 9

Energy 112 (2016) 232e240

Contents lists available at ScienceDirect

Energy
journal homepage: www.elsevier.com/locate/energy

Determinants of fuel consumption in mining trucks


Saeid R. Dindarloo a, *, Elnaz Siami-Irdemoosa b
a
b

Department of Mining and Nuclear Engineering, Missouri University of Science and Technology, Rolla, MO, USA
Department of Geosciences and Geological and Petroleum Engineering, Missouri University of Science and Technology, Rolla, MO, USA

a r t i c l e i n f o

a b s t r a c t

Article history:
Received 31 March 2016
Received in revised form
13 June 2016
Accepted 18 June 2016
Available online 5 August 2016

Analysis of fuel consumption in a large surface mine, during more than 5000 cycles of material transportation, revealed considerable variability in the data. Truck fuel estimation based on the mining truck
manufacturers' manuals/estimates is not capable of capturing this variability in the fuel consumption
data. Partial least squares regression and autoregressive integrated moving average methods were
employed to examine the effect of cyclic activities on fuel consumption, and to recommend relevant
remedies for consumption reduction. Proper modications of the operation can result in improved cycle
times. Consequently, minimizing some cyclic activities would enhance energy efciency. The truck
empty idle time was a major contributor to unnecessary fuel consumption. Since the truck queues at
shovels are a major component of the empty idle time, decisions should be reviewed to reduce the
truck queues at loading points. Improved dispatching strategies, optimal muck pile shape and size distribution, and improved shovel/loader operator skills are effective preventive measures to minimize
truck ow bottlenecks at loading points, and thus to improve energy efciency at mines.
2016 Elsevier Ltd. All rights reserved.

Keywords:
Fuel consumption
Mining trucks
Truck empty idle time
Energy efciency

1. Introduction
According to the US Energy Information Administration, transportation sector accounted for 28% of the total country's energy
consumption in 2014 [1]. Also, according to the US Environmental
Protection Agency transportation sector accounted for 26% of the
total US greenhouse gas (GHG) emissions in 2014 [2]. Nearly onethird of the total fuel consumed and GHG emitted were due to
diesel heavy duty vehicles (HDV). Therefore, modeling HDV fuel
consumption (and GHG emissions) is of paramount importance to
both the energy regulatory sectors and environmental protection
agencies. As a result, a considerable amount of literature has been
published on modeling/analyzing vehicle fuel consumption. In
recent years, the vehicle energy modeling research has been mainly
focused on light duty vehicles. For instance, Du et al. [3] performed
life cycle assessment analysis for the intensive aluminum applications in the Chinese light vehicle industry. The authors propose that
both GHG emissions and energy consumption could be further
reduced when the Al content in an automobile increases. Muratori
et al. [4] used simulation to estimate energy consumption for personal transportation in the Unitized states. Khayyam and Bab-

* Corresponding author.
E-mail address: srd5zb@mst.edu (S.R. Dindarloo).
http://dx.doi.org/10.1016/j.energy.2016.06.085
0360-5442/ 2016 Elsevier Ltd. All rights reserved.

Hadiashar [5] propose an intelligent energy management system,


using fuzzy logic controllers, for light duty hybrid electric vehicles.
Hao et al. [6] evaluated the effects of the new Chinese policies on
energy consumption and GHG emissions of passenger cars. However, there are several recent publications on the specic issue of
HDV energy modeling that are further discussed in the following.
Whyte et al. [7] analyzed historical fuel consumption data, and
modeled freight truck energy demand in Ireland. Zamboni et al. [8]
analyzed fuel consumption and GHG emissions data of HDVs in
urban areas of Genova. Zhao et al. [9] propose that improvements in
engine efciency, aerodynamic drag, and rolling resistance of
certain types of trucks in China can enhance the truck energy efciency. Zamboni et al. [10] studied the experimental speed patterns of HDVs in urban and port areas, and related the speed
patterns to vehicle fuel consumption and GHG emissions. Liu et al.
[11] analyzed GHG emissions of diesel buses in Beijing using data
collected by a portable emission measurement system. The authors
observed that there were considerable discrepancies between the
actual measured data and engine certications. The results of this
study emphasize the importance of building energy analyzing
models that are based on actual working conditions. Kousoulidou
et al. [12] evaluated different GHG emission calculation models and
vehicle simulation tools, and compared the results of experimental
studies with those of simulations. It was revealed that energy
consumption estimation using simulation might not be accurate

S.R. Dindarloo, E. Siami-Irdemoosa / Energy 112 (2016) 232e240

enough in the case of complex powertrain architectures because


much more input information is required than the conventional
powertrains. More recently, Fontaras et al. [13] experimentally
evaluated the methodology for monitoring and certication of CO2
emissions from HDVs in Europe, which is under preparation by the
European Commission. In this study, both the simulation results for
estimating fuel consumption and GHG emissions from the simulation were compared with actual on-road measurements. In the
simulations, between 2 and 4% deviations from the actual measurements were observed.
On the other hand, material handling in surface mining operations differs from other industries that use heavy vehicles. A mining
material loading and haulage cycle includes the major steps of
loading, hauling, dumping, maneuvering, and returning to the
loading points. Mining roads have harsher surface conditions than
highways, with grades up to 15%. Payloads can exceed 350 tonnes,
and the amount of dust generated is usually higher. Moreover, the
operation is characterized by shorter cycles than transportation in
other industries. These mining haulage conditions result in unique
trends associated with fuel consumption, which necessitate a
customized study.
Among the few research studies published on the subject,
Kecojevic and Komljenovic [14] studied the effects of power and
engine load factors on fuel consumption. Sahoo et al. [15] proposed a generic model to benchmark energy consumption for
dump trucks in surface mines. Dessureault [16] used regression
and articial neural networks to predict both diesel fuel and
electricity consumption. Bogunovic et al. [17] developed an integrated data environment system for analysis of energy consumption in a surface coal mine. Kecojevic et al. [18] proposed a
methodology to estimate the energy consumption in a surface
bituminous coal mine. However, the effect of cyclic activities/
times on haul truck fuel consumption per cycle has not been
addressed in the literature. In a recent study by the authors of this
paper [19] the relationship between mining truck fuel consumption and cyclic activities was modeled within a neural network
approach. However, because of the black-box nature of the neural
networks, no functional relationship between the independent
variables (e.g., loading, hauling, etc.) and fuel consumption was
built. This study builds upon the previous publication, and contributes to this area by proposing a functional relationship between the variables. Also, fuel consumption rates for different
truck operating modes are determined. Furthermore, in this study,
the results are compared with the truck manufacturer's recommended formula.
In summary, this research study aims to predict fuel consumption per cycle of operation of heavy mining dump trucks at a
selected mining operation, which is based on cyclic loadinghauling-dumping activities. The results could be used to estimate
fuel costs for alternative scenarios, so that a scenario could be
selected that reduces fuel consumption and greenhouse gas emissions. The study applies partial least squares regression (PLSR), and
autoregressive integrated moving average (ARIMA) to predict fuel
consumption based on cyclic activities (e.g., loading time, loaded
haulage time, etc.).
2. Methods
Since the effects of cyclic activities on truck fuel consumption in
surface mining operations have not been addressed in the literature, a methodology for assessing these effects is proposed in this
study. In this approach, the amount of fuel consumed per one full
cycle of material handling relates to the cycle components. A full
cycle is composed of the major steps of loading, maneuvering,
hauling, dumping and returning. Each of the cycle components is

233

associated with a specic portion of the total fuel consumed per


one full cycle of material handling. Therefore, a functional relationship between the cycle components (as the independent variables) and the amount of fuel consumed per cycle (as the
dependent variable) determines the share of each component to
the overall truck fuel consumption.
Although truck manufactures propose rough estimates for fuel
consumption for different truck models and sizes, there are
considerable uncertainties in actual fuel consumption rates,
which are due to specic conditions of different mine sites. In
other words, mine specic conditions (road quality, grades, etc.)
result in considerable deviations in actual fuel consumption with
respect to the manufacturer's estimates. Under/over estimation of
a mine's energy requirement results in considerable difculties in
deciding on the mine budget. In order to provide reliable and sitespecic truck fuel consumption estimations, in this study, historical data of material handling operations are collected and
analyzed. In this approach, rstly the variabilities in fuel consumption, which are due to the specic mine site conditions are
taken into the account. This objective is achieved through using a
sufciently large database of historical fuel consumption data.
The importance of using experimental (on-road) fuel consumption measurements in the accurate energy modeling is highlighted in the literature. For instance, Lie et al. [11] presents the
results of modeling the on-board GHG measurements from urban
buses in the Beijing metropolitan. Kousoulidou et al. [12] evaluated the performance of several emission models and vehicle
simulation tools, and highlighted the importance of accurate real
measurements in models/simulations performances. Secondly, a
functional relationship between the cycle components and total
fuel consumption (per cycle) is determined, which identies the
share of each component to the overall fuel consumption.
Therefore, specic components can be targeted to be modied in
order to reduce fuel consumption. In order to achieve the
mentioned objectives, methods of partial least square regression,
and autoregressive integrated moving average are introduced in
sections 2.1 and 2.2.
2.1. Partial least squares regression
The PLSR method is mainly used for modeling linear regression
between multiple dependent variables and (multiple) independent
variables. An advantage of this method over the ordinary linear
multiple regression (MR) is that PLSR combines the basic functions
of regressing models, principal components analysis, and canonical
correlation analysis [21]. In the context of linear regressions, the
least-squares solution for equation (1) is given by equation (2).

Y XB

(1)

1

XT Y
B XT X

(2)

Often, the problem is that XTX is singular, either because the


number of variables (columns) in X exceeds the number of objects
(rows), or because of collinearities. PLSR circumvents this by
decomposing X into orthogonal scores (T) and loadings (P) (equation (3)) [22]:

X TP

(3)

Further, PLSR regresses Y, not on X, but on the rst a columns of


the scores. The goal of PLSR is to incorporate information on both X
and Y in the denition of the scores and loadings. The scores and
loadings are chosen in such a way to describe as much as possible of
the covariance between X and Y.

234

S.R. Dindarloo, E. Siami-Irdemoosa / Energy 112 (2016) 232e240

2.2. ARIMA

Table 2
Descriptive statistics of fuel consumption.

The fuel consumption data were analyzed and modeled via the
ARIMA technique within the context of time series analysis. Data
preparation/pre-processing is one of the rst and most important
steps in the time series analysis. The basic statistics of the fuel
consumption data are summarized in Table 2.
Developed by Box and Jenkins [23], ARIMA models provide a
statistically robust approach to time series forecasting. ARIMA
models aim to describe autocorrelations in the data [24]. In an
ARIMA model, the future value of a variable is supposed to be a
linear combination of the past values and errors, expressed as
equation (4):

Statistic

Fuel consumption (L/Cycle)

No. of observations
Minimum
Maximum
1st quartile
Median
3rd quartile
Mean
Variance (ne1)
Standard deviation (ne1)

4978
4.700
59.600
12.300
13.200
15.100
13.844
11.301
3.362

yt w0 41 yt1 42 yt2 4p ytp t  w1 t1  w2 t2


  wq tq
(4)
Where yt is the actual value, t is the random error at time t, 4i and
wj are coefcients, and p and q are integers that are often referred to
as autoregressive and moving average polynomials, respectively.
For example, the ARIMA (1,0,1) model can be represented as
equation (5):

yt w0 41 yt1 t  w1 t1

(5)

3. Results
Vital information management system (VIMS, Caterpillar Inc.)
data were obtained for more than 5000 haulage cycles at a strip
coal mine. The study trucks were CAT 785C 136-t rigid frame haul
trucks, equipped with on-board data logging systems and loaded by
a Hitachi EX1900 hydraulic shovel (11 m3 dipper) during overburden removal operations. The database included the material
handling activities, payloads, and fuel consumption per cycle.
Table 1 summarizes the descriptive statistics of all variables
after elimination of the outliers by the Grubbs test [20]. As the
result of data cleaning, 149 data points were rejected from the
initial database that contained 5, 127 records. Independent
(explanatory) variables were payload, loading time, loaded idle
time, loaded travel time, empty travel time, and empty idle time.
The dependent (output) variable was the volume of fuel consumed
per cycle. The PLSR and ARIMA techniques were used to model and
predict fuel consumption based on the vital information management system data per Table 1.
The result of applying the PLSR technique in predicting the fuel
consumption for 100 cycles is illustrated in Fig. 1. One hundred cycles
(2%) of the available 5000 cycles were randomly selected and used for
testing the model. In order to examine the model generalizability,
these cycles were not used in the model building step.

Fig. 1. Actual vs. PLSR-predicted fuel consumptions for 100 cycles.

The root mean square error (RMSE) and mean absolute percentage error (MAPE), for the testing dataset, were 1.17 and 6.01%,
respectively. MAPE values less than 10% are associated with
excellent model performance in modeling the unseen data.
On the other hand, the ARIMA model is a data-oriented
approach that is adapted from the structure of the data [25].
Therefore, forecasting is based on a linear combination of past
observations that needs a stationary series without any specic
trend in data. A widely used method of transferring data into a
stationary series is application of differencing. In most cases, one or
two orders of differencing are enough to prepare data for the
method. Parameter d is the differencing order in the combined
model (i.e., ARIMA (p,d,q)). Application of ARIMA to seasonal data
needs further differencing in the seasonal part. In this case, the
model is called seasonal ARIMA (SARIMA) and is represented by
SARIMA (p,d,q)(P,D,Q)S with the seasonal differencing order of D
and cycle of S. P and Q are the autoregressive and moving average
components of the seasonal part of the data. In this case, D 1 was
selected to deal with the seasonality of the model. Best p, q, P, and Q
parameters of the SARIMA model were obtained based on both the

Table 1
Descriptive statistics of the variables associated with fuel consumption of mining dump trucks [19].
Variables
Independent
Payload (tonnes)
Loading time (sec)
Loaded idle time (sec)
Loaded travel time (sec)
Empty travel time (sec)
Empty idle time (sec)
Dependent
Fuel consumed (L/Cycle)

Minimum

Maximum

Mean

Standard deviation

Coefcient of variation

51
30
29
21
18
0

164
899
854
687
649
7200

124.01
226.86
45.77
162.15
119.88
137.13

10.57
52.35
26.85
36.08
41.64
328.69

0.09
0.23
0.59
0.22
0.35
2.40

13.84

3.36

0.24

4.7

59.6

S.R. Dindarloo, E. Siami-Irdemoosa / Energy 112 (2016) 232e240

corrected Akaike [26] information criterion and Schwarz Bayesian


criterion [27] as SARIMA (1,1,1)(0,1,0)S100.
The low RMSE and MAPE for the 100-cycle prediction period
show the good performance of the SARIMA model in predicting fuel
consumption (Table 3).
Model parameters along with the upper and lower 95% condence intervals for predictions are summarized in Table 4. Prediction of fuel consumption for 100 cycles mimics the behavior of the
curve, with relatively narrow 95% bounds, as depicted in Fig. 2.
Model diagnostics is the last part of SARIMA modeling that can
be performed through inspection of residuals. The model residuals
were examined through the Ljung and Box test [28] for satisfaction
of the white noise series condition. A white noise series is a
sequence of statistically independent and identically distributed
random variables.
Other important ARIMA modeling features, such as autocorrelation, partial autocorrelation, normality, trend, homogeneity, and
stationarity tests are discussed in the Appendix section.
4. Discussion
Analysis of the experimental fuel consumption data revealed
that the truck empty idle time is a major contributor to unnecessary
fuel consumptions. Since truck queues at shovels are a key
component of empty idle time, efforts should be made to reduce
the queues. The major reasons of increased empty idle time and
hence energy consumption are lack of a proper working face and
oor, non-optimal rock size distribution/heap location, and bottlenecks in dispatching and truck allocation strategies. In any case,
elimination of the underlying cause(s) not only reduces energy
costs but it also increases production rates [29]. In addition to
technical parameters, truck/shovel operator skills and driving styles
have direct effects on energy consumption. Causal studies are
needed to quantify the effect of workforce-related parameters on
energy consumption.
The variables showing the highest positive correlation with fuel
consumption are empty travel time, loaded travel time, and empty
idle time (Table 5). Both the loaded and empty travel times depend
on the haulage distances and maximum allowable speeds. The
haulage road surface condition, grades, and curves determine the
maximum achievable speed. Thus, better road conditions have a
considerable impact on fuel consumption by reducing both loaded
travel time and empty travel time.
The third variable that is strongly positively correlated with fuel
consumption is empty idle time. The operating trucks are either
idle in shovel and crusher queues or are hauling/dumping materials. In a typical surface mining operation, the majority of idle time
while empty is spent waiting in the shovel queues for the following
reasons:
1.
2.
3.
4.
5.

two or more empty trucks arrive together at a loading point;


the working face is not clean/ready for the loading operation;
oversize crushed rock increases the digging time;
the shovel operator's lack of skill; and
poor weather conditions.

Therefore, appropriate remedies can be implemented to reduce


the empty idle time, and improve fuel consumption and production

Table 3
Goodness of t statistics for the SARIMA
model.
RMSE
MAPE

2.244
4.753%

235

Table 4
SARIMA model parameters.
Parameter

Value

Lower bound (95%)

Upper bound (95%)

AR(1)
MA(1)

0.716
0.970

0.688
0.982

0.743
0.958

rates. For example:


 if two or more empty trucks arrive at the same loading point
during a very short time span, the dispatching and truck allocation strategy can be modied to reduce the truck empty idle
times, which are due to forming a truck queue in a loading
point;
For instance, Ercelebi et al. [30] and Dindarloo et al. [29]
demonstrated the effect of an optimized dispatching program in
minimizing truck idle times in two separate studies. Ercelebi et al.
[30] employed a closed queening network algorithm to optimize
both the number of trucks and truck dispatching in Orhaneli open
pit coal mine in turkey. As a result, truck queue lengths in the
loading points, and thus the truck empty idle times were reduced
considerably (over 30%) in the case study mine. Dindarloo et al. [29]
propose a stochastic simulation framework for optimizing material
handling in open pit mines using the discrete event simulation
method. The model was examined in Golegohar iron ore mine in
Iran, and resulted in a considerable reduction (40e44%) in truck
idle times.
 if there is a lack of proper material for shovels to dig and load,
appropriate auxiliary machinery, such as dozers, can be used;
 if the materials are oversize, the drilling patterns and blasting
operations can be modied to produce both an optimal rock size
distribution and heap displacement.
For instance, Amiel [31] reports a nearly 5% reduction in truck
loading times that was achieved through improved rock fragmentation at Tarkawa gold mine in Ghana. This reduction in the loading
operation time reects in reduced truck idle times, and thus less
energy consumption.
 if the shovel operator is not competent enough to load trucks in
minimum passes/time, proper actions including technical
training are recommended.
For instance, Patnayak et al. [32] conducted a shovel performance monitoring study in two oil sand mines in Canada. This
study showed that shovel performance was signicantly inuenced
by the operators. Up to 50% variation in the shovel productivity was
observed between different operators who had different skill levels.
Therefore, a less skilled (trained) operator reduces the loading
operation productivity signicantly, which consequently causes
increased truck idle times.
These deciencies in operations have been extensively evaluated in the literature in terms of production rates and costs (see, e.g.
Ref. [29]) The current study demonstrated another major benet of
minimizing the empty idle time: considerable reduction in fuel
consumption.

4.1. Rate of fuel consumption


The PLSR regression model is presented in equation (6). The
coefcients of the time variables in equation (6) are equal to the
rate of fuel consumption, in different operating modes, in Liter/

236

S.R. Dindarloo, E. Siami-Irdemoosa / Energy 112 (2016) 232e240

Fig. 2. ARIMA model, validation and prediction data, and observed data.

Table 5
Correlation matrix for the six independent variables and one dependent variable [19].
Variable

Payload

Loading time

Loaded idle time

Loaded travel time

Empty travel time

Empty idle time

Fuel consumption

Payload
Loading time
Loaded idle time
Loaded travel time
Empty travel time
Empty idle time
Fuel consumption

1.000
0.079
0.004
0.259
0.107
0.015
0.203

0.079
1.000
0.037
0.066
0.007
0.084
0.050

0.004
0.037
1.000
0.020
0.047
0.141
0.150

0.259
0.066
0.020
1.000
0.584
0.007
0.663

0.107
0.007
0.047
0.584
1.000
0.326
0.668

0.015
0.084
0.141
0.007
0.326
1.000
0.538

0.203
0.050
0.150
0.663
0.668
0.538
1.000

second/cycle. The associated consumption rates are presented in


Liter/second/1000 cycles of operation in Table 6. PL is in metric
tonnes, and all other independent variables in equation (6) are in
seconds.


F


L
1:37071 0:00483  PL 0:00398  LT 0:00499
cycle
 ES 0:01471  ETR 0:00278  LS 0:0519
 LTR
(6)

Where.

Table 6
The rates of truck fuel consumption in different operations.
Operating mode

Fuel consumption rate


L/sec/1000 cycles

Loading
Empty idle
Empty traveling
Loaded idle
Loaded traveling

3.98
4.99
14.71
2.78
51.9

F: Fuel consumption (liters per cycle);


PL: Payload (metric tonnes);
LT: loading time (seconds);
ES: Empty idle time (seconds);
ETR: Empty travel time (seconds);
LS: Loaded idle time (seconds), and:
LTR: Loaded travel time (seconds).
Expectedly, the rate of fuel consumption in the loaded traveling
mode is the highest (51.9 L/sec/1000 cycles), which is followed by
the empty hauling mode (14.7 L/sec/1000 cycles). The difference
between the rates of consumption in these two modes is considerable (37.19 L/sec/1000 cycles), which is due to the signicantly
higher engine power requirements for hauling the load uphill than
for trucks returning empty downhill. In other words, only a one
second increase in the hauling time results in 37.19 L more fuel
consumption in the loaded haulage operation (in every 1000 cycles) than the empty travel operation (for the same distance). The
rate of consumption of the other three operating modes of empty
idle, loading, and loaded idle are 4.99, 3.98, and 2.78 L/sec/1000
cycles, respectively, that are considerably lower than those of the
other two modes. In other words, Table 6, represents the effect of
increasing (decreasing) each independent variable by 1 s in the
amount of more (less) fuel consumption in liters per 1000 cycles.

S.R. Dindarloo, E. Siami-Irdemoosa / Energy 112 (2016) 232e240

For instance, reducing the average truck empty idle time from 137 s
to 136 s results in saving nearly 5 L of fuel per every 1000 cycles per
each truck (see Tables 1 and 6).
4.2. Truck manufacture estimates
The truck manufactures provide estimates for truck fuel consumption per hour for different engine powers and load factors
(LF). A truck engine LF is the portion of a truck's maximum available
power that is required to do a specic operation. For instance, LF for
loaded hauling is equivalent to the power portion needed to haul a
specic amount of load in a specic distance under operational
constraints such as, road surface conditions, speed, operator driving
skills and style, weather condition, road grades, etc.
According to Caterpillar, depending on the operation conditions
a typical LF may range between 0.2 and 0.5. The equipment manufacturers use LF to roughly estimate fuel consumption in different
operating modes. In this study, actual fuel consumption data of a
eet of trucks were used. Therefore, the problem in this study was
determining the relationship between cycle times and fuel consumption. One advantage of the proposed model in this paper is
that the estimates are more accurate and site-specic than those
proposed by manufacturers, which are based on one independent
variable of LF for different engine powers/types. Understandably,
the purpose of manufacturers' formula is to provide a very rough
estimate of fuel consumption.
According to the Caterpillar performance handbook [33] a CAT
785 C (1450 HP) truck's hourly fuel consumption is estimated at
53.7e80.6 L/h, 80.6e107.7 L/h, and 107.5e134.4 L/h for the low,
medium, and high LF levels, respectively.
The three LF levels are described as in the following:
 Low: Continuous operation at an average gross weight less than
the recommended level (excellent haul roads, and no
overloading).
 Medium: Continuous operation at an average gross weight
approaching the recommended level (minimal overloading and
good haul roads).
 High: Continuous operation at or above maximum recommended gross weight (overloading and poor haul roads).
Load Factor Guide (average engine load factor based on the
application description for each range):
 Low 20%e30%.
 Medium 30%e40%.
 High 40%e50%.
Using the above information, Table 7 summarizes the calculation of fuel consumption in liter per second for different operating
modes, which are based on the manufacturer's manual.
The amount of fuel consumed in each cycle is calculated using
the recorded cyclic times (in seconds) and the last two columns of
Table 7 for both the lower bound (minimum) and upper bound
(maximum) estimates. The MAPE values obtained using the
Caterpillar estimates are compared with those of the proposed
model in this study, and are summarized in Table 8.
Table 8, shows a considerably better performance of the new
model than the LF model (i.e., the manufacturer estimates based on
load factor levels) in estimating actual fuel consumptions per cycle
in terms of the MAPE metric. Twenty ve cycles are randomly
selected to depict the difference between the two models in terms
of fuel consumption predictability in Fig. 3.
Fig. 3 shows that the actual fuel consumptions generally lay
between the minimum and maximum estimates of the

237

manufacturer. Nevertheless, there are a few cycles wherein the


actual fuel consumption has been less than the minimum estimate
of the manufacture. The range between the lower and upper bound
estimates is signicant. The average actual fuel consumption for the
25 cycles in Fig. 3 is 13.1 L/Cycle, while the average manufacturer's
lower and upper bound estimates are 12.8 L/Cycle and 17.9 L/Cycle,
respectively. Though the actual fuel consumption lay in the range
[12.8, 17.9], the difference between the lower and upper bounds is
about 40%. A forty percent uncertainty translates into signicant
difculties in deciding a mine's operational budget. On the other
hand, the new model presents a very close estimate of the actual
data, with an average of 12.6 L/Cycle (3.8% underestimation).
Therefore, the need for building an experimental model of fuel
consumption based on historical data is justied in order to be able
to accurately estimate the location of the fuel consumption curve in
the wide upper-lower range.
It is worth mentioning that the rates of fuel consumption for the
truck idle operating modes are signicantly smaller than those of
hauling modes (see Table 6). This is due to fact that when a truck is
idle the corresponding engine load factor is minimal (see Table 7).
Therefore, the amounts of fuel consumed in liter per second of the
idle operations are considerably lower than those of the hauling
(empty and loaded) operations, which are associated with signicantly higher engine loads. Thus, expectedly, one second reduction
in idle modes (e.g., empty idle time) results in smaller reduction in
energy consumption than hauling operations. However, it should
be taken into the account that the empty and loaded idle times are
unnecessary fuel consuming modes that should be minimized
(ideally eliminated).
In order to further compare the models' performances, Figs. 4e6
respectively illustrate coefcients of determination for the new
model estimations and the Min/Max of manufacture estimates
versus the actual fuel consumptions for the whole dataset. The
coefcient of determination(R2) obtained by plotting the scattergram of the new model versus actual fuel consumptions (Fig. 4)
shows a considerably higher linear correlation between the two
variables (R2 0.76) than the other two models (Figs. 5e6), which
are based on the manufacturer's estimates (R2 0.53 and 0.51,
respectively).
It should be noted that, the case study operation was concentrated in a specied region of the mine during the year that the data
were collected. Although the cycles are relatively short (see
Table 1), the proposed method can be examined in the case of
longer haulage distances as well. With this regard, equation (6) can
be modied to take into the account the dynamic proles (i.e., other
haulage distances):
In equation (6), the empty and loaded travel times (ETR and LTR)
are the only distance (prole)-dependent variables, while the other
four independent variables (i.e., payload, loading time, empty idle
time, and loaded idle time) do not depend on haulage distances.
Therefore, equation (6) can be readily modied as in equation (7):


F


L
1:37071 0:00483  PL tonnes 0:00398
cycle
 LTSec: 0:00499  ESSec: 0:01471
 3:6

HL m

 0:00278  LSSec:
ALS km=h

0:0519  3:6

HL m


AES km=h

(7)

In equation (7), HL is the haulage distance in meters, ALS is the


average loaded hauling speed in km/h (also, can be obtained from

238

S.R. Dindarloo, E. Siami-Irdemoosa / Energy 112 (2016) 232e240

Table 7
Calculation of the estimated fuel consumption (per second) in different operating modes, and for different LF levels.
Operating Mode

Engine load factor

Minimum fuel consumption

Maximum fuel consumption

Liter/Sec.
Loading
Loaded idle
Loaded hauling
Empty hauling
Empty idle

Low (0.2e0.3)
Low (0.2e0.3)
High (0.4e0.5)
Medium (0.3e0.4)
Low (0.2e0.3)

Table 8
Comparing the new model's performance with the manufacturer's estimates (whole dataset).
Model

MAPE (%)

Min. Manufacturer
Max. Manufacturer
New Model

15.5
21.4
7.7

the manufacture's performance handbook), AES is the average


empty returning speed in km/h, and 3.6 is a unit conversion factor.
All other variables have the same meanings and units as in equation

0.01492
0.01492
0.02992
0.02239
0.01492

0.02239
0.02239
0.03733
0.02992
0.02239

(6). Therefore, equation (7) can be used for different (longer)


haulage distances.
Although possible eet interferences, road networks/gradients,
and seasonal conditions have not been directly considered in
equations (6) and (7), their effects are highly correlated with cycle
times. Since the database contains the haulage operational data
during one year, the effects of distance changes, seasonal variations,
and network alterations (during one year) are already incorporated
in the database. However, these variations are representative of one
year of operation of the specic case study.
In summary, although the methods were used in a specic case
study, they are applicable in other operations within a same

Fig. 3. Comparing the actual fuel consumptions with the new model and the manufacturer's estimates for the randomly selected 25 cycles.

Fig. 4. Correlation between the actual fuel consumption and new model's estimates.

Fig. 5. Correlation between the actual fuel consumption and minimum manufacturer's
estimates.

S.R. Dindarloo, E. Siami-Irdemoosa / Energy 112 (2016) 232e240

239

g(k), is dened in equation (A.1):

gk

5. Conclusions
Estimating the mining truck's fuel consumption using PLSR and
ARIMA techniques provided practical insights into the major causes
of the unnecessary and excessive fuel consumption. PLSR was used
to build a functional relationship between the cyclic activities and
the amount of fuel consumed per cycle. Therefore, each activity's
share of the total consumed fuel was identied. On the other hand,
in ARIMA (similar to the manufacturer's formula) only one variable
(total fuel consumption) was analyzed within the context of time
series analysis. In ARIMA, no relationship between the variables
was constructed, but the (fuel consumption) data variability with
time and its predictability were evaluated. Evaluation of the effect
of different independent variables (cyclic activities) on fuel consumption revealed the considerable effect of truck idle times at
shovels in the overall fuel consumption per cycle. Improved dispatching strategies, optimal muck pile shape and size distribution,
and improved shovel/loader operator skills are effective preventive
measures to minimize truck ow bottlenecks at loading points.
Thus, the overall haulage eet empty idle times can be decreased,
which results in more energy efciency and cost effective haulage
operations. Furthermore, estimating fuel consumption, according
to Caterpillar's guidelines for CAT 785 C expectedly resulted in a
wide uncertainty margin, which justies the need for building sitespecic experimental models of fuel consumption.
Appendix
A central feature in the development of time series models is an
assumption of some form of statistical equilibrium. A particular
assumption of this kind is that of stationarity. A stationary time
series can be usefully described by its mean, variance, and autocorrelation function [23]. Autocorrelation is the internal correlation
of the observations in a time series, usually expressed as a function
of the time lag between observations. The autocorrelation at lag k,

1.0

0.8

Autocorrelation

framework. Any truck-shovel analysis depends on haulage distances. The haulage distances can be very dynamic depending on
the operation. This study showed the (inter) dependencies of cycles
and fuel consumption. Since the underlying principals are the same
for long and short proles, it is expected that similar results will be
obtained in the case of longer roads, albeit this proposition needs to
be examined in separate studies.
It is also worth mentioning that the proposed models tend to
underestimate fuel consumption by a factor of 4e6% (see Fig. 1),
which needs to be considered in the mine operational budget.

(A.1)

EXt  m2

Where, E denotes the expected value, Xt, t 0, 1, 2, represents


the values of the series, and m is the mean of the series.
A plot of the sample values of the autocorrelation against the lag
is known as the autocorrelation function or correlogram, and is a
basic tool in the analysis of time series, particularly for indicating
possibly suitable models for the series [34]. The autocorrelograms
of the fuel consumption data (see Table 2) are shown in Fig. A.1. The
suspension bridge pattern in Fig. A.1a is a typical characteristic of
non-stationary and seasonal time series. On the other hand, partial
autocorrelations (PACF) measure the degree of association between
various lags when the effects of other lags are eliminated (Fig. A.1b).
Both the ACF and PACF are primary graphical tools for inspecting
the time series, and deciding on the order of the autoregressive and
moving average components.
Normality, trend, homogeneity, and stationarity tests are
important statistical metrics in evaluation of time series. In summary, the results showed that the fuel consumption time series was
not normal, there was a trend in the data, and the series were nonhomogenous and non-stationary.

0.6

0.4

0.2

0.0
0

-0.2

8 10 12 14 16 18 20 22 24 26 28 30 32 34 36

Lag

1.0

Partial autocorrelation

Fig. 6. Correlation between the actual fuel consumption and maximum manufacturer's estimates.

EXt  mXtk  m

0.8

0.6

0.4

0.2

0.0
0
-0.2

8 10 12 14 16 18 20 22 24 26 28 30 32 34 36

Lag

Fig. A.1Autocorrelation (A) and partial autocorrelation (B) diagrams.

References
[1] EIA. 2015. Monthly energy review. US Energy Information Administration;
March 2015.
[2] EPA. U.S. Greenhouse gas inventory report: 1990-2014. US Environmental
Protection Agency; 2015.

240

S.R. Dindarloo, E. Siami-Irdemoosa / Energy 112 (2016) 232e240

[3] Du JD, Han WJ, Peng YH, Gu CC. Potential for reducing GHG emissions and
energy consumption from implementing the aluminum intensive vehicle eet
in China. Energy 2010;35(12):4671e8.
[4] Muratori M, Moran MJ, Serra E, Rizzoni G. Highly-resolved modeling of personal transportation energy consumption in the United States. Energy
2013;58:168e77.
[5] Khayyam H, Bab-Hadiashar A. Adaptive intelligent energy management system of plug-in hybrid electric vehicle. Energy 2014;69:319e35.
[6] Hao H, Liu Z, Zhao F, Li W, Hang W. Scenario analysis of energy consumption
and greenhouse gas emissions from China's passenger vehicles. Energy
2015;91:151e9.
 Gallacho
 ir BP. Modelling HGV freight transport energy
[7] Whyte K, Daly HE, O
demand in Ireland and the impacts of the property construction bubble. Energy 2013;50(1):245e51.
 M, Carraro C, Marelli S, Capobianco M.
[8] Zamboni G, Malfettani S, Andre
Assessment of heavy-duty vehicle activities, fuel consumption and exhaust
emissions in port areas. Appl Energy 2013;111:921e9.
[9] Zhao H, Burke A, Miller M. Analysis of Class 8 truck technologies for their fuel
savings and economics. Transp Res Part D Transp Environ 2013;23:55e63.
 M, Roveda A, Capobianco M. Experimental evaluation of
[10] Zamboni G, Andre
heavy duty vehicle speed patterns in urban and port areas and estimation of
their fuel consumption and exhaust emissions. Transp Res Part D Transp
Environ 2015;35:1e10.
[11] Liu Z, Ge Y, Johnson KC, Shah AN, Tan J, Wang C, et al. Real-world operation
conditions and on-road emissions of Beijing diesel buses measured by using
portable emission measurement system and electric low-pressure impactor.
Sci Total Environ 2011;409(8):1476e80.
[12] Kousoulidou M, Fontaras G, Lonza L, Dilara P. Overview of emission and trafc
models and evaluation of vehicle simulation tools. JRC Publ 2013;85667. DOI:
10.2790/99031 (print), 10.2790/98977 (online), http://publications.jrc.ec.
europa.eu/repository/handle/JRC85667.
[13] Fontaras G, Grigoratos T, Savvidis D, Anagnostopoulos K, Luz R, Rexeis M, et al.
An experimental evaluation of the methodology proposed for the monitoring
and certication of CO2 emissions from heavy-duty vehicles in Europe. Energy
2016;102:354e64.
[14] Kecojevic V, Komljenovic D. Haul truck fuel consumption and CO2 emission
under various engine load conditions. Min Eng 2010;62(12):44e8.
[15] Sahoo LK, Bandyopadhyay S, Banerjee R. Benchmarking energy consumption
for dump trucks in mines. Appl Energy 2014;113:1382e96.

[16] Dessureault S. Data mining, mining data: energy consumption modelling. CIM
Bull 2007;100:1e7.
[17] Bogunovic D, Kecojevic V, Lund V, Heger M, Mongeon P. Analysis of energy
consumption in surface coal mining. SME Trans 2009;326:79e87.
[18] Kecojevic V, Vukotic I, Komljenovic D. Production, consumption and cost of
energy for surface mining of bituminous coal. Min Eng 2014;66(1):51e7.
[19] Siami-Irdemoosa E, Dindarloo SR. Prediction of fuel consumption of mining
dump trucks: a neural networks approach. Appl Energy 2015;151:77e84.
[20] Grubbs FE. Sample criteria for testing outlying observations. Ann Math Stat
1950;21(1):27e58.
[21] Zhang M, Mu H, Li G, Ning Y. Forecasting the transport energy demand based
on PLSR method in China. Energy 2009;34(9):1396e400.
[22] Mevik B-H, Wehrens R. The pls package: principal component and partial
least squares regression in R. J Stat Softw 2007;18(2):1e23.
[23] Box GEP, Jenkins G. Time series analysis, forecasting and control. San Francisco, CA: Holden-Day; 1970.
[24] Hyndman RJ, Athanasopoulos G. Forecasting: principles and practice. OTexts
publishers; 2013. Retrieved from, http://otexts.org/fpp/.
[25] Pai P-F, Lin C-S. A hybrid ARIMA and support vector machines model in stock
price forecasting. Omega 2005;33(6):497e505.
[26] Akaike H. A new look at the statistical model identication. IEEE Trans
Automatic Control 1974;19(6):716e23.
[27] Schwarz GE. Estimating the dimension of a model. Ann Stat 1978;6(2):461e4.
[28] Ljung GM, Box GEP. On a measure of a lack of t in time series models. Biometrika 1978;65(2):297e303.
[29] Dindarloo SR, Osanloo M, Frimpong S. A stochastic simulation framework for
truck and shovel selection and sizing in open pit mines. J South Afr Inst Min
Metall 2015;115(1):1e11.
[30] Ercelebi SG, Bascetin A. Optimization of shovel-truck system for surface
mining. J South Afr Inst Min Metall 2009;109(7):433e9.
[31] Amiel A. Blast fragmentation optimization at Tarkwa gold mine using 6 Sigma
methodologies. J South Afr Inst Min Metall 2008;108(11):669e81.
[32] Patnayak S, Tannant DD, Parsons I, Del Valle V, Wong J. Operator and dipper
tooth inuence on electric shovel performance during oil sands mining. Int J
Min Reclam Environ 2008;22(2):120e45.
[33] Handbook. Caterpillar performance. forty-fth ed. Peoria, IL: Caterpillar. Inc.;
2015.
[34] Everitt BS. Cambridge dictionary of statistics. West Nyack, NY: Cambridge
University Press; 2002.

You might also like