A data-based investigation of vehicle maintenance cost components using ANN

Abstract

Vehicle maintenance is a critical and vital operational component that determines vehicle performance and service longevity. The severity of vehicle usage has been defined as one of the key factors that determine vehicle maintenance requirements. Nigeria is a tropical country with intense heat, poor road network and quality. These factors severely affect car performance and raises maintenance demands toward ensuring vehicle reliability and optimum performance. In this study, vehicle maintenance cost and fuel consumption data in terms of cost and volume, together with the mileage coverage as the vehicle usage data, for two corporate organizations were analyzed. The data was collected and systematically analyzed. The common faults were categorized and their frequency of occurrence was determined. An artificial Neural Network (ANN) model was developed for predicting future maintenance cost, given a set of anticipated vehicle usage inputs. The model has an overall correlation R-value of 0.76645.

Full text

IOP Conference Series: Materials Science and Engineering
PAPER • OPEN ACCESS
A data-based investigation of vehicle maintenance cost components
using ANN
To cite this article: Aderibigbe Israel Adekitan et al 2018 IOP Conf. Ser.: Mater. Sci. Eng. 413 012009
View the article online for updates and enhancements.
This content was downloaded from IP address 165.73.192.9 on 10/09/2018 at 19:31

1
Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd
1234567890‘’“”
ICESW IOP Publishing
IOP Conf. Series: Materials Science and Engineering 413 (2018) 012009 doi:10.1088/1757-899X/413/1/012009
A data-based investigation of vehicle maintenance cost
components using ANN
Aderibigbe Israel Adekitan1,*, Adetokun Bukola2 and Okokpujie Kennedy1
1Department of Electrical and Information Engineering, College of Engineering,
2Department of Electrical Engineering, Pan African University of Basic Sciences,
Technology and Innovation, hosted at Jomo Kenyatta University of Agriculture and
Abstract. Vehicle maintenance is a critical and vital operational component that determines
vehicle performance and service longevity. The severity of vehicle usage has been defined as
one of the key factors that determine vehicle maintenance requirements. Nigeria is a tropical
country with intense heat, poor road network and quality. These factors severely affect car
performance and raises maintenance demands toward ensuring vehicle reliability and optimum
performance. In this study, vehicle maintenance cost and fuel consumption data in terms of cost
and volume, together with the mileage coverage as the vehicle usage data, for two corporate
organizations were analyzed. The data was collected and systematically analyzed. The common
faults were categorized and their frequency of occurrence was determined. An artificial Neural
Network (ANN) model was developed for predicting future maintenance cost, given a set of
anticipated vehicle usage inputs. The model has an overall correlation R-value of 0.76645.
Keywords: ANN model, vehicle maintenance, predictive cost analysis, data pattern
recognition, road transportation, statistics
1. Introduction
Supply chain reliability is the back bone of many commercial operations, and it also ensures adequate
distribution and availability of various products, supplies and equipment for both domestic and industrial
uses. Product and material distribution can be via air, water, rail and road transportation. Transportation
is vital for socio-economic interactions. Freight transportation must be cost effective, efficient, safe, fast
and reliable to ensure itch-free national distribution of goods and economic development. According to
[1], road transportation is the most active mode for haulage transportation. It was estimated that 75% of
the total 2200 billion tonne-kilometres of freight transportation in Europe was via road. In the Nigerian
transport industry, the road transport sub-sector contributes about 90% to the Gross Domestic Product.
As at the end of 2017, the estimated number of vehicles in Nigeria was 11,583,331 [2, 3].
The transportation industry in Nigeria is plagued by many challenges due to poor planning, poor road
network and condition, scarce funds, corruption and defective policies, insufficient vehicle fleets, and so
forth. Vehicles daily ply poorly constructed, and poorly maintained roads with pot holes and many
unpaved road sections which create inherent hazards [4]. This negatively impacts vehicle reliability,
resulting in repetitive vehicle breakdowns with an attendant increase in maintenance cost. To ensure
vehicle availability for service, corporations need to dedicate a sizable amount of their budget towards
vehicle maintenance and overhaul. Most of the vehicles in Nigeria were imported. It is only in recent
times, due to new Government policies that both indigenous and foreign car assembling plants are
springing up again in the country,
after decades of zero production due to various economic challenges.
Foreign Exchange (FOREX) issue is a recurrent problem in Nigeria due to FOREX scarcity and price
fluctuations. Currently, one US dollar is around 365 Naira. This makes vehicle importation very
expensive, and as a result of the high cost, most of the cars and trucks imported into the country are
fairly-used, and they are rarely brand new. This implies that a good percentage of the service life of the
vehicle is used up before arrival in Nigeria, and upon arrival into the country, deterioration sets in due to
many factors and realities of the Nigerian society. The only way to ensure vehicle reliability and prolong
Covenant University, Ota, Nigeria
Technology, Nairobi, Kenya
*E-mail address: ade_kitan@yahoo.com

2
1234567890‘’“”
ICESW IOP Publishing
IOP Conf. Series: Materials Science and Engineering 413 (2018) 012009 doi:10.1088/1757-899X/413/1/012009
vehicle service life is through adequate and timely vehicle maintenance [5], but this is often challenged
by the need to manage scarce resources.
A visit to the car pack and maintenance warehouse of some organizations will reveal run-down and
abandoned vehicles, with the tyres of some of the vehicles already removed, and the chassis supported by
stone slabs. A number of times, this occurs as a result of scarce spare part, low maintenance budget and
excessive car breakdowns leading to abandonment. Vehicles abandoned in this way end up deteriorating
further with time, and this increases the cost of restoration, and reduces the sale value, if the vehicle is
sold off in that state. If such cars are parked in open and unfenced locations in front of factories without
adequate security, the car parts end up being stolen by scavengers. Cases of stolen head lights, tyres,
batteries and so forth, have been reported.
In maintenance engineering and management, a number of maintenance components and parameters can
be extracted for objective analysis [6]. This study presents and examines the maintenance data of two
corporate organizations for fault frequency trending and for data parameter relationship identification.
An ANN model was developed using the fuel consumption, cost of fuel, mileage covered and the
maintenance cost data for the sampled period as model parameters.
2. Case study: haulage company truck and corporate pool car maintenance
The effect of the harsh economic conditions and poor infrastructural development is quite severe on
haulage company vehicles that need to move bulk load from one point to another [7]. The trucks are the
lifeline of the companies and must be operational to deliver value. According to the study in 2009 by
AAA Car Care, more than 62% of vehicles are operated on severe service conditions, some of these
conditions include:
▪ Driving in hot weather under frequent stop and start traffic
▪ Driving on dusty or muddy roads and on roads with gravel spread
▪ Driving below 50 miles per hour over long distances
▪ Transporting heavy loads or towing a trailer
Nigeria is a tropical country, with an average temperature of about 28C, which could rise to about 35C
in the northern region. When haulage trucks travel between local communities, a number of unpaved
roads will be encountered. Most of these roads are unpaved, dusty and muddy, and the pot holes in some
road sections are filled with stones and pebbles. Poor roads and insufficient road network creates heavy
traffic along major roads, and this induces stop and start vehicle movement. Since there is no dedicated
lane for trucks and bulk vehicles, heavy loaded trucks caught up in traffic are also forced to stop and start
with the slow moving traffic. By these realities, most vehicles in Nigeria are operating under severe
conditions, and this requires frequent maintenance and oil change to limit damage, and prolong vehicle
service life. Frequent maintenance has an associated cost implication which may result in non-
compliance in a bid to reduce expenses. In Europe, road transport accounts for 17.5% of total greenhouse
gas emissions [8], and this will be aggravated in developing countries by poorly maintained vehicles with
sooty vehicle exhausts.
A. The haulage company truck maintenance analysis
Commercial trucks travel thousands of kilometres with heavy loads every year, and this journey is most
times under severe conditions. Prior to embarking on a journey, trucks ought to meet the minimum safety
and maintenance requirement to prevent sudden breakdown en route to their destination. This is often not
the case; a number of broken down trucks can be seen on major highways and on interstate routes in

3
1234567890‘’“”
ICESW IOP Publishing
IOP Conf. Series: Materials Science and Engineering 413 (2018) 012009 doi:10.1088/1757-899X/413/1/012009
Nigeria. Some of these loaded trucks are sometimes at the breakdown spots for weeks, with the driver
and driver assistant sleeping in, or around the truck to protect the cargo until the truck is fixed. Often the
needed technical assistance, funds and or spare parts is coordinated from the office of the logistic
company which is in one state, while the broken down truck is in another. The repair process is
sometimes slow and this can lead to cargo theft and product damage for perishable goods.
Fig.1 Faulty components and their frequency of occurrence
Fig.2 The contribution of each fault and repairs to total maintenance cost
Figure 1 shows the fault log and their frequency of occurrence for 15 haulage trucks, for one operational
year. Figure 2 shows the contribution of each fault and repairs to the total maintenance cost, based on the
average cost of each, expressed in percentage. The availability for service of each of the truck throughout
the year could not be established via records. It is possible that some of the trucks were out of service for
extensive periods within the year. The data set shows that the gas filter, oil filter and the water separator
were replaced 22 times, while engine oil replacement was done 34 times. For the 15 trucks, if oil and
filter replacements were done at least once in 6 months, the filters would have been replace 30 times.
This indicates a possible cost saving measure by topping the engine oil without a complete replacement
of the filtration system, and this could negatively impact the engine performance of heavy duty trucks.
This could also imply that some of the trucks as earlier suggested were out of service for certain periods,
and as such there was no need to perform some periodic maintenance.

4
1234567890‘’“”
ICESW IOP Publishing
IOP Conf. Series: Materials Science and Engineering 413 (2018) 012009 doi:10.1088/1757-899X/413/1/012009
Depending on the type and design, the number of trailer truck tyres can be 10 and up to 18 per truck.
Substandard tyres are gradually flooding the market, though they are sometimes cheaper but they do not
last long, and are usually replaced more frequently. Poor Nigerian roads, heavy loading of the trucks and
the spread of low quality tyres are likely factors that may have resulted in the replacement of 36 tyres
representing 37.97% of the total maintenance expenses.
B. The corporate pool car maintenance analysis
Pool cars are vital in corporate environment for conveying personnel from one place to another for
various business purposes. In some organization, each car is assigned to a specific unit or department, but
the maintenance of all the cars is often managed by the maintenance department. This study analyses the
maintenance and fuel related data for 25 pool cars of a corporate organization. From the plot in Figure 3
and Figure 4, periodic maintenance (filter and oil replacement) occurred 16 times and this is followed by
brake related issues that occurred 14 times. Labour charges for various services contributed significantly
to the maintenance cost profile. The highest maintenance cost is 15.6397% of the total maintenance cost,
and it was incurred for replacing a single propeller shaft. This emphasises how the nature of the fault and
the affected component determines maintenance expenses.
Fig.3 Frequency of occurrence of various car faults
For the 5 months of maintenance cost data for the pool cars, a single tyre was not replaced, this is in
contrast to the haulage truck scenario. This could be due to the significant difference in car and truck
loading or it could imply that some tyres were recently replaced outside of the 5 months window data.
For the cars, the frequency of brake related issues is quite high as compared with that of the trucks. Since
the two organizations are in different states, this could be an indication of a more intense stop-and-start
traffic in the state of the pool car organization.
Traffic and vehicle statistics can be applied for various studies and analysis. In [9], crash data was
collected and analysed for acquiring valuable information about traffic safety. Maintenance planning and
management can also be improved by providing valuable statistics about past trends and by showing the
relationships between various maintenance parameters. This can help in making informed management
decisions. This study carried out prediction analysis for the pool car data using the ANN model in Figure
7. The model was implemented in MATLAB.

5
1234567890‘’“”
ICESW IOP Publishing
IOP Conf. Series: Materials Science and Engineering 413 (2018) 012009 doi:10.1088/1757-899X/413/1/012009
Fig.4 Total car maintenance cost breakdown
3. Maintenance cost prediction using ANN
ANN has found application in various fields of life for data cluster analysis, prediction and so forth. In
the study by [10], neural network was applied for vehicle maintenance scheduling. In this study, the goal
of the pattern recognition and relationship identification among the parameters is to determine the extent
to which the collected vehicle usage parameters can be used to predict the vehicle maintenance cost of
the pool cars, and this is achieved using an Artificial Neural Network (ANN) model. The ANN model is
deployed using the Nonlinear Input-Output time series tool box in MATLAB. The model is a dynamic
neural network, which include tapped delay lines. Through dynamic filtering, the ANN model uses past
values of the six vehicle usage parameters, that is, the Fuel Cost (FC) in Nigerian Naira, the Fuel Volume
in litres, the car Mileage in km, the Normalized Fuel Cost, the Normalized Fuel Volume, and the
Normalized Mileage as model inputs, to predict future Maintenance Cost as the model target.
Fig.5 Percentage monthly contribution to the total maintenance cost for 5 consecutive months
The ANN model comprises the inputs, the hidden layer with delays, and the output layer. The hidden
layer has 2 delays and 20 neurons with sigmoid activation functions. The analysed data is for a period of
5 months, comprising 111 days of vehicle usage data. The 111 sample dataset is divided into three sets,
in the ratio 70:15:15 for training, validation and testing. The ANN network training was performed using
the Levenberg-Marquardt back-propagation algorithm, and the performance after the implementation
was analysed using Regression Analysis and the Mean Squared Error (MSE).

6
1234567890‘’“”
ICESW IOP Publishing
IOP Conf. Series: Materials Science and Engineering 413 (2018) 012009 doi:10.1088/1757-899X/413/1/012009
The monthly contribution to the total expenditure on fuel and maintenance is shown by the pie charts of
Figure 5, while Figure 6 shows the variation in the four key data parameters across the months
Fig.6 Percentage variation of the 4 study data components across the months
Fig.7 The neural network
4. Results and Discussion

7
1234567890‘’“”
ICESW IOP Publishing
IOP Conf. Series: Materials Science and Engineering 413 (2018) 012009 doi:10.1088/1757-899X/413/1/012009
The performance of the model can be analysed using the MSE and the correlation coefficient (R-value)
of the regression, which indicates how close the model output prediction is, to the target (maintenance
cost). As shown in Figure 8, for the training data set, the R-value is 0.76833, it is 0.76893 for the
validation, and 0.80774 for the test. The overall R-value for the three analyses is 0.76645. As shown in
Figure 8, the dashed lines in the regression plot represent a perfect line of fit, when the output is exactly
equal to the target; this is an ideal situation. The solid lines in the regression plot stands for the real
regression line showing the true relationship between the target and the predicted maintenance cost. The
R-value can be increased by using more data, say for a period of two years in order to increase the
training dataset. The best validation performance of the model as measured using the MSE occurred at
epoch (iteration) 7, as shown in Figure 9. The MSE gradually reduced from the peak value until the
model execution was terminated after 6 consecutive validation failure checks. Figure 10 shows the
variation of the gradient coefficient, the Marquardt adjustment parameter (Mu), the total validation fails
and the number of iterations.
Fig.8 Regression plot
Fig.9 ANN model validation performance plot using mean squared error

8
1234567890‘’“”
ICESW IOP Publishing
IOP Conf. Series: Materials Science and Engineering 413 (2018) 012009 doi:10.1088/1757-899X/413/1/012009
Figure 11 shows the output of the model and the desired target (MC) for the training, validation and
model test analysis. The error histogram is shown in Figure 12, and it reveals that the maximum model
error for the maintenance cost prediction is 2397 Naira. The error is computed as the difference between
the target and the model output. . The magnitude of the error can be reduced by using more training data
to improve the accuracy of the model.
Fig.10 Network Training State
Fig.11 Output element response plot
Fig.12 Error Histogram

9
1234567890‘’“”
ICESW IOP Publishing
IOP Conf. Series: Materials Science and Engineering 413 (2018) 012009 doi:10.1088/1757-899X/413/1/012009
Using the trained ANN model, different sets of the vehicle usage parameters were supplied as inputs to
the model, and the model predicted the equivalent maintenance cost in Naira. This implies that given an
anticipated vehicle mileage, fuel consumption in litres and fuel cost over a period of time, the company
can predict how much will be spent on maintaining the pool cars, thereby making this ANN model a
useful tool for maintenance budgeting.
5. Conclusion
Maintenance expenses makes up a sizable portion of an organization’s budget, and as such it must be
cost effectively managed by making informed decisions. Vehicles are vital to corporate business, either
as a direct source of income (haulage) or for business support (pool cars). In this study, the maintenance
and vehicle fuel consumption operational data of two corporate organizations were collected, studied and
analysed. The different types of vehicle faults that make up the maintenance cost components were
identified together with their frequency of occurrence. An ANN model was developed using the pool car
usage data as input to predict the car maintenance cost. The result shows that there is a significant
correlation between the predictor inputs, and the predicted maintenance cost, using a set of fuel volume,
fuel cost and car mileage input data. This model can provide useful information for vehicle maintenance
cost budget planning. The scope of this research can be extended by collecting other parameters, both
qualitative and quantitative to enable the implementation of an improved prediction model.
Acknowledgment
The Authors appreciate Covenant University Centre for Research, Innovation and Discovery for
sponsoring our participation at the conference.
References
[1] Eurostat. (2017). Freight Transport Statistics. Available:
http://ec.europa.eu/eurostat/statistics-explained/index.php/Freight_transport_statistics
[2] NBS. (2017, Transport Statistics.
Available: http://www.nigerianstat.gov.ng/pdfuploads/TRANSPORT.pdf
[3] NBS. (2018, Road Transport Data (Q4 2017).
Available: http://www.nigerianstat.gov.ng/pdfuploads/Road_Transport_Data_-
_Q4_2017_.pdf
[4] A. Adekitan, "Root Cause Analysis of a Jet Fuel Tanker Accident," International Journal of
Applied Engineering Research, vol. 12, pp. 14974-14983, 2017.
[5] E. Baffour-Awuah, "Service Quality in the Motor Vehicle Maintenance and Repair Industry:
A Documentary Review."
[6] C. Bolu, Modeling maintenance productivity measurement of engineering production
systems: Discrete event simulation approach vol. 13, 2013.
[7] A. O. Adewumi and O. J. Adeleke, "A New Model for Optimizing Waste Disposal Based on
Customers’ Time Windows and Road Attributes," Applied Mathematical Sciences, vol. 10,
pp. 2051-2063, 2016.
[8] EEA. (2011). Most car makers must further improve carbon efficiency by 2015 Available:
https://www.eea.europa.eu/highlights/most-carmakers-must-further-improve/
[9] Y. Liu, Z. Li, J. Liu, and H. Patel, "Vehicular crash data used to rank intersections by injury
crash frequency and severity," Data in Brief, vol. 8, pp. 930-933, 2016/09/01/ 2016.
[10] S. Kamlu and V. Laxmi, "An effective method for maintenance scheduling of vehicles using
neural network," in Lecture Notes in Electrical Engineering vol. 453, ed, 2018, pp. 51-66.