Anda di halaman 1dari 6

International Journal of Mining Science and Technology 23 (2013) 441–446

Contents lists available at SciVerse ScienceDirect

International Journal of Mining Science and Technology


journal homepage: www.elsevier.com/locate/ijmst

Statistical approach to determination of overhaul and maintenance cost


of loading equipment in surface mining
Lashgari Ali a,⇑, Sayadi Ahmad Reza b
a
Department of Mining Engineering, West Virginia University, Morgantown, WV 26506, USA
b
Department of Mining, Faculty of Engineering, Tarbiat Modares University, Tehran 14115-111, Iran

a r t i c l e i n f o a b s t r a c t

Article history: The purpose of this research was to develop a new approach in determination of overhaul and mainte-
Received 2 September 2012 nance cost of loading equipment in surface mining. Two statistical models including univariate exponen-
Received in revised form 21 October 2012 tial regression (UER) and multivariate linear regression (MLR) were used in this study. Loading
Accepted 25 December 2012
equipment parameters such as bucket capacity, machine weight, engine power, boom length, digging
Available online 30 May 2013
depth, and dumping height were considered as variables. The results obtained by models and mean abso-
lute error rate indicate that these models can be applied as the useful tool in determination of overhaul
Keywords:
and maintenance cost of loading equipment. The results of this study can be used by the decision-makers
Overhaul and maintenance cost
Loading equipment
for the specific surface mining operations.
Surface mining Ó 2013 Published by Elsevier B.V. on behalf of China University of Mining & Technology.
Univariate exponential regression
Multivariate linear regression
Principal component analysis

1. Introduction the equipment [4]. US Army Corps of Engineers suggests includ-


ing some factors to account for regional price variations and infla-
Principal loading equipment in surface mining includes electric tion [5]. Nichols described two different methods to predict
shovels, hydraulic front and backhoe shovels, and wheel loaders. hourly maintenance cost: (i) as a percentage of equipment pur-
Overhaul and maintenance cost (parts and labor) of loading equip- chased price and (ii) as a percentage of the straight-line depreci-
ment represents a significant percentage of the overall operating ation of the equipment [6]. Nunnally uses term ‘‘repair cost’’ and
cost. Data provided by the InfoMine indicate that overhaul and suggests that hourly cost should take into the account the current
maintenance cost is represented by 32%, 50%, 59%, and 64% of total year in the life of the equipment, the total number of years in the
operating cost for wheel loader, backhoe shovel, hydraulic front economic life of the equipment, the lifetime repair cost (which is
shovel, and cable shovel, respectively [1]. expressed as a percentage of original purchasing cost), and the to-
Overhaul cost is associated with scheduled reconstruction and/ tal number of hours that equipment is expected to operate during
or replacement of major components such as engines and trans- its lifetime [7]. This model assumes that the accumulated repair
missions. Maintenance cost is related to unscheduled repairs cost at the end of each year would be the same, regardless of
and scheduled service of both minor and major components, number of hours the equipment has been used. Model developed
excluding overhaul activities. The cost is often expressed in one by the InfoMine uses the following variables in determination of
of three ways: (i) as a constant cost per hour, (ii) as a constant hourly maintenance and overhaul cost: capital cost of the equip-
cost per unit production, or (iii) as a percentage of the equip- ment, so-called ‘‘experience-based factors’’, the typical number of
ment’s purchasing cost at a particular point of the equipment life operating hours per year, and overhaul and repair labor hourly
[2]. According to Cox, estimation of equipment overhaul and wages and benefits [1].
maintenance cost can be expressed as a percentage of purchasing This paper presents a novel method for determination of over-
cost modified by the multiplying factors of types of services [3]. haul and maintenance costs for principal loading equipment in
Caterpillar indicates that overhaul and maintenance costs of surface mining. Two statistical models are used including univar-
mining equipment are affected by its application, operating iate exponential regression (UER) and multivariate linear regres-
conditions, ownership period, maintenance practices, and age of sion (MLR). The former considers only one equipment
parameter, while the latter accounts for multiple parameters as
⇑ Corresponding author. Tel.: +1 3047779760. the predictor variables in determination of overhaul and mainte-
E-mail address: allashgari@mix.wvu.edu (A. Lashgari). nance cost.

2095-2686/$ - see front matter Ó 2013 Published by Elsevier B.V. on behalf of China University of Mining & Technology.
http://dx.doi.org/10.1016/j.ijmst.2013.05.002
442 A. Lashgari, A.R. Sayadi / International Journal of Mining Science and Technology 23 (2013) 441–446

2. Methods where yij is the value of the standardized matrix; xij the value of the
jth column and ith row in the matrix Xnm; and  xj the mean value of
Technical and cost data of loading equipment are collected from the jth column.
the InfoMine [1]. Fig. 1 shows the technical parameters (variables) The correlation coefficients matrix R of data samples can be set
for loading equipment. Each of the parameters listed in Fig. 1 is up for each of independent variables according to yij, using the fol-
used as a predictor variable in the UER model while all parameters lowing equation:
related to each type of equipment are simultaneously used as the
R ¼ ½1=ðn  1Þ  Y 0 Y ð2Þ
predictor variables in MLR model.
0
A quick estimate of overall overhaul and maintenance cost can where Y is the standardized data matrix; and Y the transpose of
be conducted by the UER model. This model can be applied in an standardized data matrix.
initial phase of the project evaluation in which a sufficient equip- In the subsequent step, the matrix of eigenvalue kj and eigen-
ment data are available and if the mine decision-makers prefer vector cj of the matrix R can be derived. Eigenvectors were used
to have a quick tool to evaluate different alternatives. However, to obtain unrotated factor loadings. Rotated loading values in prin-
this model does not provide a high accuracy in cost determination. cipal components indicate the contribution of each variable to the
The overall overhaul and maintenance cost (Y) can be expressed in PC. The loading value l is given by following equation:
the following form: Y = aXm, where Y is in $/h, X represents the spe- X
m pffiffiffiffi
cific equipment parameter, while a and m are constants deter- lj ¼ kj  cj ð3Þ
mined by regression analysis. j¼1

The MLR model can be very useful if the mine decision-makers


Two major approaches for selection of significant PCs can be con-
have sufficient data related to specific equipment parameters. In
sidered. The eigenvalue-one criterion, also known as the Kaiser cri-
comparison with UER model, MLR uses several parameters simul-
terion, is the simplest and the most common method used to solve
taneously, therefore it is usually assumed to be more accurate. This
the number of principal component problems. With respect to this
model is expressed in the following form: Y = b + a1X1 + a2X2 +  +
criterion, PCs with eigenvalue greater than one, that are considered
aiXi, where Y is the overhaul and maintenance cost (in $/h), X1 to Xi
to be the most significant, are selected as independent variables
are equipment parameters, and b, a1 to ai are constants determined
[11]. On the other hand, all of PCs are selected as independent vari-
by the regression analysis.
ables in MLR model and t-test is used to select the significant inde-
One of the greatest challenges in MLR model is multicollinearity
pendent variables. In this research, the latter approach was used and
phenomena in which predictor variables have a high correlation
MLR was applied to all of PCs. It allows obtaining the linear
with each other. The high correlation among predictor variables
functions as relationships between significant PCs, and therefore
can lead to unstable regression model so that the variance and
the overhaul and maintenance cost. The established functions are
covariance of coefficients are very high. In this research, MLR mod-
transformed and the final cost can be obtained as function of the
el was applied at three stages. Firstly, with a respect to omitting
initial specific equipment parameters. In this transformation,
correlation among independent variables, the principal component
eigenvector of correlation matrix obtained by the PCA is multiplied
analysis (PCA) can be used [8]. Secondly, a significant principal
by the coefficients determined from MLR. The relationships, ob-
components (PCs), obtained by PCA, were used as independent
tained by this approach, were used as the standardized functions.
variables in MLR model, and finally at third stage, the relationship
The final coefficients are calculated by following equation:
between dependent variable (Y) and original predictor variables
(X1 to Xi) was established. Fig. 2 shows a schematic view of MLR X ¼ ðX dl  X ave Þ=Sx ð4Þ
model that is applied in this research.
where X is the final regression coefficient; Xdl the regression coeffi-
PCA is a dimension reduction technique in which the goal is to
cient obtained by standardized data sample; and Xave and Sx mean
find a few principal components that explain a large proportion of
and standard deviation of the variable, respectively.
the total sample variance of the variables while the PCs have no
correlation with each other. Using PCA, the original variables are
3. Results and discussion
transformed into linear combinations called principal components
[9]. Principal components are often used as input to other analysis
Table 1 shows the basic statistical parameters (minimum, max-
such as regression.
imum, average, and standard deviation) of loading equipment
Suppose that the data samples named Xnm are collected. The m is
parameters such as bucket capacity, weight, engine power, dump
the number of variables and n is the number of data samples. In order
height, digging depth, and boom length. Data on overall overhaul
to eliminate the effects of grade and dimension between variables, at
and maintenance costs are also shown for wheel loaders, hydraulic
first the Xnm should be standardized by the following equation [10]:
P backhoe and front shovels, and electric shovels.
xij  ð1=nÞ  ni¼1 xij Table 2 shows the results of UER model along with the
yij ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P ð1Þ
coefficients of determination, while Fig. 3 shows the overall over-
½1=ðn  1Þ  ni¼1 ðxij  xj Þ2
haul and maintenance cost based on a single loading equipment

Loading equipment

Electric shovel Hydraulic front shovel Backhoe shovel Electric shovel


Bucket capacity (BC) Bucket capacity (BC) Bucket capacity (BC) Bucket capacity (BC)
Machine weight (W) Machine weight (W) Machine weight (W) Machine weight (W)
Power (HP) Power (HP) Power (HP) Power (HP)
Boom length (BL) Dumping height (DH) Dumping height (DH) Dumping height (DH)
Digging depth (DD)

Fig. 1. Technical parameters (variables) of loading equipment.


A. Lashgari, A.R. Sayadi / International Journal of Mining Science and Technology 23 (2013) 441–446 443

... ...

Fig. 2. A schematic view of the MLR model.

Table 1
Statistical data of loading equipment.

Bucket capacity (m3) Weight (kg) Power (HP) Dump height (m) Digging depth (m) Boom length (m) Overhaul and maintenance cost ($/h)
Wheel loader
Min 1.00 6950.8 73.0 2.50 3.15
Max 40.50 262,176.4 2300.0 7.30 156.23
Average 9.89 75,282.4 650.7 4.17 40.13
St. Dev 11.13 80,289.4 671.5 1.59 47.70
Backhoe shovel
Min 0.28 8035.0 54.0 5.10 4.10 6.65
Max 12.00 192,000.0 1086.0 12.20 14.50 102.13
Average 3.69 67,771.8 407.2 7.98 8.06 35.86
St. Dev 4.08 62,621.4 320.9 2.04 2.76 33.47
Hyd. front shovel
Min 2.30 43,499.5 250.0 5.00 46.87
Max 42.80 789,250.7 3350.0 13.90 607.90
Average 15.61 272,134.3 1342.4 9.86 208.42
St. Dev 12.05 220,954.2 970.5 2.49 152.25
Electric shovel
Min 5.40 252,197.0 685.0 10.70 67.50
Max 61.20 2,041,166.0 7000.0 19.50 269.48
Average 23.92 686,701.4 2926.7 15.41 150.40
St. Dev 16.27 492,703.4 1988.2 2.75 64.57

parameter such as bucket capacity. It can be noticed from Table 2 If the overhaul and maintenance cost of the loading equipment,
and Fig. 3 that 94.29% of variation in overall overhaul and mainte- with a specific parameter (variable), is known as (A), the following
nance cost for electric shovel can be explained by bucket capacity, equation may be used for determination of the overhaul and main-
96.34% for hydraulic front shovel, 96.83% for backhoe shovel, and tenance cost with a different variable (B):
97.53% for wheel loader.
CostðBÞ ¼ CostðAÞ  ½X B =X A m ð5Þ
Table 2
where m is the scaling exponent determined by the regression; X
UER models and coefficients of determination.
the cost for a specific variable; and A and B the values of two vari-
Equation R2 ables (specific equipment parameters).
Electric shovel This exponential rule is generally satisfactory, not only for
Y = 20.902(BC)0.6358 94.29 bucket capacity, but also for other variables. Table 3 shows the m
Y = 1.2845(HP)0.6025 92.86
values for all variables used in the model.
Y = 0.0159(W)0.684 84.67
Y = 0.2723(BL)2.2876 87.69 Statistical analysis of predictor variables shows intense correla-
tions among them (Table 4). With respect to existing correlations,
Hyd. front shovel
Y = 17.071(BC)0.9131 96.34 it is essential to take this problem away from the data and then ap-
Y = 0.2364(HP)0.9408 95.32 ply MLR to uncorrelated independent variables [12]. Values given
Y = 0.004(W)0.8697 94.22 in Table 4 are often close to 1, which indicates a significant corre-
Y = 0.8166(DH)2.3433 78.52 lation between independent variables. Applying PCA on loading
Backhoe shovel equipment parameters to describe their interrelation pattern, the
Y = 14.173(BC)0.7534 96.83 number of PCs will equal the number of original independent vari-
Y = 0.1056(HP)0.9603 93.88
Y = 0.0019(W)0.8874 98.20
ables. These new variables are independent linear combinations of
Y = 0.0295(DH)3.2827 81.41 original variables and they retain the maximum possible variance
Y = 0.2134(DD)2.3313 72.16 of the initial set. Table 5 shows the eigenvectors of correlation ma-
Wheel loader trix that represents the matrix of the weights for the PCs, which
Y = 2.9702(BC)1.0714 97.53 shows the relative importance of each standardized parameter in
Y = 0.0223(HP)1.1396 97.27 the PC calculations.
Y = 0.0005(W)1.0048 97.54
As an example, the following equation indicates the relation be-
Y = 0.1726(DH)3.4589 93.75
tween PC3 and specific parameter of wheel loaders:
444 A. Lashgari, A.R. Sayadi / International Journal of Mining Science and Technology 23 (2013) 441–446

Fig. 3. Overall overhaul and maintenance cost vs. bucket capacity.

Table 3 Table 5
Scaling exponent for independent variables (m). Eigenvectors of correlation matrix.

BC DH DD BL W HP Type Variable PC1 PC2 PC3 PC4 PC5


Wheel loader 1.071 3.458 1.004 1.139 Wheel loader BC 0.499 0.508 0.691 0.127
Hydraulic shovel 0.913 2.343 0.869 0.940 DH 0.495 0.830 0.256 0.017
Cable shovel 0.635 2.287 0.684 0.602 W 0.503 0.113 0.565 0.645
Backhoe shovel 0.753 3.282 2.331 0.887 0.960 HP 0.503 0.200 0.372 0.754
Cable shovel BC 0.517 0.183 0.048 0.835
BL 0.469 0.818 0.308 0.129
PC3 ¼ 0:691  BC þ 0:256  DH  0:565  W  0:372  HP ð6Þ W 0.497 0.544 0.500 0.455
HP 0.517 0.036 0.808 0.281
Hyd. shovel BC 0.293 0.722 0.610 0.144
The high value of bucket capacity BC indicates that this parameter is DH 0.361 0.667 0.642 0.113
the most important parameter in this relationship. W 0.629 0.105 0.341 0.691
Eigenvalues of correlation matrix are shown in Table 6. The HP 0.623 0.153 0.316 0.699

eigenvalue indicates the importance of the respective factors in Backhoe BC 0.442 0.490 0.576 0.197 0.441
explaining the variation of the data. It can be seen that the eigen- DD 0.432 0.606 0.147 0.612 0.225
DH 0.447 0.459 0.137 0.727 0.206
values are arranged in decreasing order. W 0.455 0.401 0.034 0.198 0.769
In order to determine the overall overhaul and maintenance HP 0.460 0.148 0.792 0.139 0.348
cost, the analyses of the PC scores were performed by the stepwise

Table 4
Correlation matrices of independent variables.
A. Lashgari, A.R. Sayadi / International Journal of Mining Science and Technology 23 (2013) 441–446 445

Table 6 variable selection procedure. For example, the result of MLR model
Eigen values of correlation matrix. for cable shovels is shown in Table 7.
Type Variable Eigen value Total variance (%) Beta column in Table 7 indicates the standardized coefficients of
Loader PC1 3.915 97.878 regression. These coefficients are useless for estimation, but their
PC2 0.059 1.472 advantage (compared to the ones that are not standardized) is that
PC3 0.018 0.456 their magnitudes allow comparing the relative contribution of each
PC4 0.008 0.194 independent variable in the estimation. For example, with regard
C. shovel PC1 3.695 92.365 to this column, PC1 is the most important explanatory parameter
PC2 0.279 6.984 of overall overhaul and maintenance cost of cable shovels. The B
PC3 0.020 0.509
PC4 0.006 0.142
column in Table 7 indicates regression coefficient of PCs for each
function. For evaluation of significance of regression coefficients,
H. shovel PC1 2.449 61.223
PC2 1.465 36.617
the t-test was applied and the results indicate statistically signifi-
PC3 0.070 1.754 cance of explanatory parameters. It can be noticed that PC1 is sta-
PC4 0.016 0.406 tistically significant in this relationship.
Backhoe PC1 4.503 90.066 Table 8 shows the coefficient of determination for each model.
PC2 0.416 8.315 It can be seen that, in average, 96.98% of variation in overall cost
PC3 0.062 1.249 can be explained by this approach. Since the increase in number
PC4 0.018 0.367
of independent variables can result in an increase of the R2, using
PC5 0.000 0.003
the adjusted coefficient of determination (adjR2) can be more sui-
ted for comparing these models. The adjusted R2 is interpreted as
that of the R2 value except it takes into account the number of
Table 7 degrees of freedom. It is adjusted by dividing the error sum and to-
Results of MLR analysis for cable shovels. tal sums of squares through their respective degrees of freedom.
Beta B Std. Err. of B t p-Level The following equation indicates relations between R2 and adjR2:
Intercept 150.4007 4.59608 32.7237 adjR2 ¼ 1  ½ðresidual SS=dfÞ=ðtotal SS=dfÞ ð7Þ
PC1 0.963960 0.073863 32.3840 2.48140 13.0507
PC2 0.096001 0.073863 11.7289 9.02409 1.2997 where residual SS is the error sums of squares; the total SS the total
PC3 0.064401 0.073863 29.1345 33.41512 0.8719
sums of squares; and df their respective degree of freedom.
PC4 0.091198 0.073863 78.1788 63.31820 1.2347
In order to establish the relationships among independent
variables of loading equipment and overall cost, the B coefficients
in Table 7 are multiplied to eigenvectors of correlation matrix
given in Table 6. Using the Eq. (4), the variables are returned again
Table 8
to their initial positions (un-standardized) and the relationships
Coefficients of determination of each function.
between overhaul and maintenance cost and equipment
Item R2 adjR2 parameters are established. For example, relationship between
Hydraulic shovel 95.94 95.04 independent variables of cable shovels and overall overhaul and
Cable shovel 95.09 92.91 maintenance cost (OOMC) is obtained as follows:
Wheel loader 99.35 99.09
Backhoe 97.54 96.82 OOMCð$=hÞ ¼ 1:029BC þ 5:511BL þ 0:00003W þ 0:008HP
Average 96.98 95.96
 6:194 ð8Þ

Fig. 4. Comparison of the results.


446 A. Lashgari, A.R. Sayadi / International Journal of Mining Science and Technology 23 (2013) 441–446

The results for all of loading equipment can be expressed as follows: 4. Conclusions

Cable Shovel OOMC ¼ 6:194 þ 1:029BC þ 5:511BL Determination of the overhaul and maintenance cost of loading
þ 0:00003W þ 0:008HP ð9Þ equipment is a very important task in surface mining operations. A
statistical approach to determination of overhaul and maintenance
cost of loading equipment (electric and front and backhoe hydrau-
Hydraulic Shovel OOMC ¼ 9:018 þ 12:33BC
lic shovels, and wheel loaders) in surface mining has been pre-
þ 0:00026DH  0:00026W sented in this paper. Two series of cost estimation equations
þ 0:08HP ð10Þ including univariate exponential regression (UER) and multivariate
linear regression (MLR) are presented and mean absolute error rate
is used for the evaluation of the models. Results indicate that these
Wheel Loader OOMC ¼ 22:806 þ 0:738BC  10:193DH models can be applied as a useful tool in determination of overhaul
þ 0:0005W þ 0:026HP ð11Þ and maintenance cost of loading equipment.

References
Backhoe Shovel OOMC ¼ 9:285 þ 3:564BC  0:366DH
 1:221DD þ 0:0002W [1] InfoMine. InfoMine USA Inc.: Mine and mill equipment cost estimator guide;
2012.
þ 0:029HP ð12Þ [2] Lanz T, Noakes M. Cost estimation handbook for the Australian mining
industry. Parkville: Australasian Institute of Mining and Metallurgy; 1993.
[3] Cox EA. Equipment economics, handbook of heavy construction. New
where OOMC is the overall overhaul and maintenance cost; and BC,
York: McGraw Hill; 1971.
DH, BL, DD, W and HP indicate bucket capacity, dump height, boom [4] Caterpillar performance handbook. Peoria: Caterpillar; 2010.
length, digging depth, weight and horse power, respectively. [5] Construction equipment ownership and operating expense schedule, regions
The mean absolute error rate (MAER) is used for the evaluation I–XII. Washington, DC: Army Corps of Engineers; 1995.
[6] Nichols HL. Moving the earth. Greenwich: North Castle Books; 1976.
of the models, which is defined in following equation [13]: [7] Nunnally SW. Construction means and methods. New Jersey: Prentice Hall;
hX i 1993.
MAER ¼ jðOOMCe  OOMCa Þ=OOMCa j:100 =n ð13Þ [8] Liu RX, Kuang J, Gong Q, Hou XL. Principal component regression analysis with
SPSS. Comput Meth Prog Biol 2003;71(2):141–7.
[9] Koop JC. On upper limits to the difference in bias between two ratio estimates.
where OOMCe and OOMCa are estimated and actual OOMC, respec- Metrika 1962;5:145–9.
[10] Jambu M. Exploratory and multivariate data analysis. New York: Academic
tively; and n the number of data. The results obtained from the UER
Press; 1991.
(based on bucket capacity) and MLR models yield the MAER of [11] Kaiser HF. The application of electronic computers to factor analysis. Educ
15.42% and 11.01% on average, respectively (Fig. 4). This indicates Psychol Meas 1960;20:141–51.
[12] Gujarati D. Basic econometrics. New York: McGraw-Hill; 2003.
that a new model can be used as a reliable tool for the determina-
[13] Kim GH, An SH, Kang KI. Comparison of construction cost estimating models
tion of the overhaul and maintenance cost of loading equipment in based on regression analysis, neural networks, and case-based reasoning.
surface mining operations. Build Environ 2004;39:1235–42.

Anda mungkin juga menyukai