computers

Review
Importance and Applications of Robotic and
Autonomous Systems (RAS) in Railway Maintenance
Sector: A Review
Randika K. W. Vithanage * , Colin S. Harrison and Anjali K. M. DeSilva
Department of Mechanical Engineering, Glasgow Caledonian University; Glasgow G4 0BA, UK
* Correspondence: [email protected]




Received: 19 June 2019; Accepted: 25 July 2019; Published: 30 July 2019


Abstract: Maintenance, which is critical for safe, reliable, quality, and cost-effective service, plays
a dominant role in the railway industry. Therefore, this paper examines the importance and
applications of Robotic and Autonomous Systems (RAS) in railway maintenance. More than 70
research publications, which are either in practice or under investigation describing RAS developments
in the railway maintenance, are analysed. It has been found that the majority of RAS developed
are for rolling-stock maintenance, followed by railway track maintenance. Further, it has been
found that there is growing interest and demand for robotics and autonomous systems in the
railway maintenance sector, which is largely due to the increased competition, rapid expansion and
ever-increasing expenses.

Keywords: robotics; autonomous systems; railway; maintenance

1. Introduction
Maintenance can be defined as a task or series of tasks that protect or reinstate the anticipated
condition of a system, and these tasks include technical, administrative, and managerial actions
taken [1,2]. In railway industry, proper maintenance of infrastructure, rolling-stock, and other resources
are vital in providing a safer, reliable, efficient, and resilient output. In Swedish railway, about 30%
of all rail and track related incidents and accidents that took place between 1988 to 2000 were due to
maintenance related causes [3]. Additionally, it has been revealed that among 700 accidents reported
over 23 countries, 37% were due to rolling-stock faults and 36% were due to failures in infrastructure [4].
Further, by referring to Pareto graph in Figure 1a, it can be realized that maintenance contributes to
most 80% of the rail and track related accidents in Sweden. In addition, by referring to the Pareto
graph in Figure 1b, it is evident that rolling-stock and infrastructure faults contributed almost 80% of
accidents summarized by the D-RAIL FP7 project.

Computers 2019, 8, 56; doi:10.3390/computers8030056 www.mdpi.com/journal/computers

 Computers 2019, 8, 56 2 of 16
Computers 2019, 8, x FOR PEER REVIEW 2 of 16

(a) (b)
32 Figure
Figure 1.
1. (a)
(a)Rail
Railand
andtrack
trackrelated
relatedincidents
incidentsand
andaccidents
accidentsin inSwedish
Swedishrailway
railway[3]
[3]and
and(b)
(b)Railway
Railway
33 accidents
accidentsconsidered
consideredby
bycause
causein
inthe
the D-RAIL
D-RAIL FP7
FP7 project
project [4].
[4].

34 Inaddition,
In addition,railway
railwaymaintenance
maintenancetasks tasksare
arecostly,
costly, and
and poor
poor maintenance
maintenanceor orfailure
failure to
to conduct
conduct
35 preventivemaintenance
preventive maintenancewill willlead
leadto to expensive
expensive consequences
consequences [5,6].[5,6]. According
Accordingto tothe
thejoint
jointreport
reportthat
that
36 was published by the United Kingdom’s (UK’s) Department for
was published by the United Kingdom’s (UK’s) Department for Transport and the Office of RailTransport and the Office of Rail
37 RegulationNetwork,
Regulation Network,rolling
rollingstock
stockmaintenance
maintenancein inthe
thecountry
countryaccounted
accounted£0.4 £0.4bn,
bn,which
whichisis around
around
38 3% of
3% of the
the total
total railway
railway expenditures
expenditures in in 2009–2010
2009–2010 andand aa typical
typical rolling-stock
rolling-stock maintenance
maintenance wouldwould
39 costabout
cost about44% 44%ofofits
its whole
whole life
life costs
costs [7].
[7]. Based
Basedononthe
thereport
report that
that was
was published
publishedby by National
NationalAudit
Audit
40 Office, UK
Office, UK inin 2015,
2015, thethe annual
annual maintenance
maintenance expenditures
expendituresof of UK’s
UK’s rail
rail infrastructure
infrastructureexceeds
exceeds£1 £1 bn
bn
41 and in
and in 2013–2014
2013–2014 and more more than
than two-thirds
two-thirdsofof25,531
25,531employees
employeesininNetworkNetworkRail, UK
Rail, UKwere
wereplaced
placed on
42 maintenance
on maintenance related
relatedjobs [8].[8].
jobs
43 Further, the
Further, the maintenance
maintenance cost cost in
in Europe
Europe railway
railway ranges from €€ 30,000
ranges from 30,000 toto 100,000
100,000 km/year
km/year [9].
[9]. In
In
44 the UK, it has been suggested that about 20%–30% of the service life maintenance
the UK, it has been suggested that about 20%–30% of the service life maintenance cost of a passenger cost of a passenger
45 rolling-stock isisassociated
rolling-stock associated with wheelset
with maintenance
wheelset [10]. On
maintenance the other
[10]. On the hand, inadequate
other hand, maintenance
inadequate
46 will reduce the
maintenance willreliability
reduce theofreliability
services, of and this leads
services, and to delays,
this leads to train cancellations,
delays, and customer
train cancellations, and
47 dissatisfactions
customer [9]. Based[9].
dissatisfactions on Based
past data, about
on past 23%,about
data, 17%, 23%,
and 10% 17%,ofand
the 10%
UK wide
of thetrain
UKdelays were
wide train
48 taken place
delays weredue to rolling-stock
taken place duefaults, track faults, and
to rolling-stock power/signal
faults, track faults,faults,and
respectively—which
power/signal faults,could
49 have been reduced orcould
respectively—which avoided
havebybeen preventive
reducedmaintenance
or avoided by [11].
preventive maintenance [11].

50 2.2.Importance
Importanceof
ofRobotic
Roboticand
andAutonomous
AutonomousSystems
Systems(RAS)
(RAS)in
inMaintenance
Maintenance

51 Railwaycompanies
Railway companiesdrivedrivetowards
towardscontinuous
continuousimprovements
improvementsof ofexisting
existinginfrastructure,
infrastructure,fleets,
fleets,
52 resources, and processes due to increased competition and rapid expansion. For
resources, and processes due to increased competition and rapid expansion. For example, it is example, it is projected
53 to increase
projected tothe UK’s national
increase the UK’srail traffic rail
national in passenger miles by 100%
traffic in passenger milesover the next
by 100% 30 the
over years,
next resulting
30 years,in
54 an increase of 11,566 rolling-stock fleets in 2046 when compared to 2017 [12]. Therefore,
resulting in an increase of 11,566 rolling-stock fleets in 2046 when compared to 2017 [12]. Therefore, optimizing
55 inspection and
optimizing repair of
inspection therepair
and rail network, facilities,
of the rail andfacilities,
network, rolling-stocks in a cost-effective
and rolling-stocks in amanner, whilst
cost-effective
56 preserving or enhancing the safety, quality, and reliability of the service remains
manner, whilst preserving or enhancing the safety, quality, and reliability of the service remains a key challenge [13].
a
57 For instance, Office of Rail and Road, UK expects Network Rail to carry out its maintenance
key challenge [13]. For instance, Office of Rail and Road, UK expects Network Rail to carry out its 17% more
58 efficiently by 17%
maintenance the end
moreof 2019. Further,
efficiently bytothe
reachendefficiency
of 2019.targets,
Further, theto
Network Rail plans targets,
reach efficiency to reducethe its
59 maintenance staff by 8% over the 2014–2019 control period [8].
Network Rail plans to reduce its maintenance staff by 8% over the 2014–2019 control period [8].
60 On the other hand, the RAS are becoming economical and feasible with recent manufacturing
61 and technological advancements. For example, over the period of 15 years from 1990, the mean
62 quality adjust price of industrial robots in the UK, United States (US), and European Union (EU) has
 Computers 2019, 8, 56 3 of 16

On the other hand, the RAS are becoming economical and feasible with recent manufacturing and
technological
Computers 2019, advancements.
8, x FOR PEER REVIEWFor example, over the period of 15 years from 1990, the mean3 quality
of 16
adjust price of industrial robots in the UK, United States (US), and European Union (EU) has dropped
63 dropped
down down 80%
by nearly by nearly 80% [14]—shown
[14]—shown in Figurein2a.
Figure 2a. According
According to the World
to the World Robotics
Robotics 2014 data,
2014 data, annual
64 annual shipments of industrial robots have increased over time, despite the drastic drop
shipments of industrial robots have increased over time, despite the drastic drop in 2009, which in 2009,was
65 which was due to world’s economic recession, see Figure 2b. Further, as shown in Figure 2b, it can
due to world’s economic recession, see Figure 2b. Further, as shown in Figure 2b, it can be seen that the
66 be seen that the same trend exists within the UK [2]. Therefore, the introduction of RAS in railway
same trend exists within the UK [2]. Therefore, the introduction of RAS in railway maintenance would
67 maintenance would be an ideal solution for achieving expected cost benefits. Atherton et al. have
be an ideal solution for achieving expected cost benefits. Atherton et al. have conducted an in-depth
68 conducted an in-depth financial and technical study to evaluate the feasibility of introducing RAS
financial and technical study to evaluate the feasibility of introducing RAS into rolling stock fluid
69 into rolling stock fluid service tasks. Here, they have concluded that an autonomous approach will
service tasks. Here, they have concluded that an autonomous approach will double the throughput of
70 double the throughput of fluid service task for only about 15% of the cost of an equivalent manual
71 fluid service
servicing task for
facility only about 15% of the cost of an equivalent manual servicing facility [15].
[15].

(a) (b)
72 Figure2.2.(a)
Figure (a)Price
Priceof
ofindustrial
industrial robots
robots over
over time
time[1]
[1]and
and(b)
(b)population
populationinin
thethe
United Kingdom
United (UK)
Kingdom (UK)
73 and annual shipments of industrial
and annual shipments of industrial robots from 1995 to 2011 [2].
from 1995 to 2011 [2].

74 InIn addition,railway
addition, railwaymaintenance
maintenancetechnicians
techniciansrisk riskown
ownsafety
safetytotoassure
assurethethesafety
safetyofofpassengers
passengersand
75 and to keep trains operating. In most occasions, these railway employees
to keep trains operating. In most occasions, these railway employees perform their jobs in unfavourableperform their jobs in
76 unfavourable
and unergonomic andenvironments.
unergonomic environments. For instance,
For instance, railway trackrailway
maintenancetrack maintenance
technicians and technicians
engineers
77 and engineers that are constantly exposed to severe weather conditions
that are constantly exposed to severe weather conditions and encounter risks while working on and encounter risks while
active
78 working on active railway tracks with live power transmission lines/rails in
railway tracks with live power transmission lines/rails in the vicinity, trains travelling at high speeds the vicinity, trains
79 travelling
from at tracks,
adjacent high speeds
workingfromat adjacent
heights, tracks, working at heights,
or in unergonomic posturesorfor in prolonged
unergonomic postures
periods for In
[16,17].
80 prolonged periods [16,17]. In addition, the workers in rolling-stock maintenance and service within
addition, the workers in rolling-stock maintenance and service within depots face a similar amount of
81 depots face a similar amount of risks. For example, in the UK, the cab front cleaning is manually
risks. For example, in the UK, the cab front cleaning is manually conducted, which exposes workers to
82 conducted, which exposes workers to highly dampened environments close to high voltage
highly dampened environments close to high voltage overhead power lines or electrified third rails [18].
83 overhead power lines or electrified third rails [18]. Another hazardous rolling stock maintenance
Another hazardous rolling stock maintenance task is transmission fluid changing process, where
84 task is transmission fluid changing process, where technicians need to execute the bulk of work
technicians need to execute the bulk of work beneath rolling-stock in unergonomic environments,
85 beneath rolling-stock in unergonomic environments, while getting exposed to hazards, such as oil
86 while
spillgetting
slip/tripexposed
risks ortoworking
hazards,toosuch as to
close oilelectrified
spill slip/trip
thirdrisks
railsor[14].
working too close
Therefore, to electrified of
the introduction third
87 rails [14]. Therefore, the introduction of RAS to such dull, dirty, and dangerous
RAS to such dull, dirty, and dangerous maintenance tasks could potentially minimize the health and maintenance tasks
88 could
safetypotentially minimize the health and safety risks on technicians.
risks on technicians.
89 The
Themajority
majorityofofrailway
railway maintenance
maintenance taskstasks are
are monotonous
monotonousand andpossibly
possibly depend
depend onon attitude
attitude
90 and
andself-discipline
self-discipline of of
technician
techniciantowards
towardsthe the
job [19]. When
job [19]. considering
When Japanese
considering high-speed
Japanese railway
high-speed
91 track maintenance—which involves a train driver and a supervisor
railway track maintenance—which involves a train driver and a supervisor who perform well who perform well defined simple
92 series of tasks
defined simpleover a prolonged
series of tasks overperiod, where it
a prolonged is easy
period, to lose
where it isfocus
easy and motivation
to lose focus andto accomplish
motivation
93 antoimmaculate
accomplish job an immaculate
[19]. Besides,job [19].
human Besides,
errorshuman errors are Such
are inevitable. inevitable.
human Such human
errors errors
in the in
railway
94 the railway include disassembly errors, inspection errors, assembly errors, decision-making
include disassembly errors, inspection errors, assembly errors, decision-making errors, and installation errors,
95 and installation
errors. Human errors errors.canHuman
causeerrors can cause
disastrous disastrous
failures, failures, and subsequently
and subsequently the loss of livesthe loss
[20].of In
96 lives [20]. In contrast, present-day robots and autonomous systems are renowned for their high
97 accuracy and path repeatability. Therefore, railway companies should consider such capabilities and
 Computers 2019, 8, 56 4 of 16

contrast, present-day robots and autonomous systems are renowned for their high accuracy and path
Computers 2019, 8,
repeatability. x FOR PEER
Therefore, REVIEW
railway 4 of 16
companies should consider such capabilities and utilize RAS to handle
monotonous or non-value-added maintenance tasks and utilize railway technicians to do advanced,
98 utilize RAS to handle monotonous or non-value-added maintenance tasks and utilize railway
challenging, and value-added tasks.
99 technicians to do advanced, challenging, and value-added tasks.
Further, it has been revealed that rolling stock maintenance could be clustered and distributed
100 Further, it has been revealed that rolling stock maintenance could be clustered and distributed
over local contractors. This will encourage them to be specialists in the field, and apply techniques
101 over local contractors. This will encourage them to be specialists in the field, and apply techniques to
to achieve increased productivity through automation [21]. Additionally, it has been identified that
102 achieve increased productivity through automation [21]. Additionally, it has been identified that
increased automation is a key objective in the Rail Technical Strategy 2012, and it is expected to obtain
103 increased automation is a key objective in the Rail Technical Strategy 2012, and it is expected to
operational cost benefits by intelligent maintenance techniques [22,23]. Therefore, the introduction of
104 obtain operational cost benefits by intelligent maintenance techniques [22,23]. Therefore, the
autonomous systems in order to eliminate or minimize human intervention in maintenance processes
105 introduction of autonomous systems in order to eliminate or minimize human intervention in
would be an ideal technique, and it can be viewed that the railway industry in the UK, including many
106 maintenance processes would be an ideal technique, and it can be viewed that the railway industry
other countries, is in the right stage to invest in automation [24,25].
107 in the UK, including many other countries, is in the right stage to invest in automation [24,25].
3. RAS in Railway Maintenance
108 3. RAS in Railway Maintenance
For decades, researchers have conducted extensive studies to introduce RAS into railway
109 For decades,
maintenance tasks. researchers
For example,have conducted
in 1987, Martlandextensive
analysed studies to introduce
how robotics could affectRAS theinto railway
locomotive
110 maintenance
rebuilding tasks. For
at Conrail’s example,
Juniata in 1987,
locomotive Martland
rebuilding shop, analysed
where he has how takenrobotics
more thancouldtwoaffect
dozenthe of
111 locomotive rebuilding at Conrail's Juniata locomotive rebuilding shop, where
processes into consideration. Through this study, he has concluded that there is only limited potential he has taken more
112 than
for two dozen
improving theofproductivity
processes into of consideration.
the rail workshop Through this RAS
through study,and he ithas
willconcluded that there
remain modest untilis
113 only limited potential for improving the productivity of the rail workshop
equipment and technology become economical [26]. In contrast, another feasibility study that was through RAS and it will
114 remain modest
conducted until equipment
by Wiercienski and Leekand twotechnology
years later has become economical
concluded that the[26].
use of Inrobots
contrast, another
to clean the
115 feasibility study
undersides that was
of Toronto subwayconducted by Wiercienski
cars is viable, and Leek
both technically two
and years later has
economically. Theyconcluded that the
have proposed a
116 use of robots to clean the undersides of Toronto subway cars is viable,
system that includes three industrial robots and an industrial vision system to wash undersides of two both technically and
117 economically.
subway They have proposed
cars simultaneously [27]. a system that includes three industrial robots and an industrial
118 vision
In system to wash
this paper, moreundersides of two subway
than 65 authentic researchcars simultaneously
articles [27]. to distinct research and
that were related
119 In this paper, more than 65 authentic research articles
developments of RAS in railway maintenance domain have been summarized that were related to distinct research
and analysed. It and
has
120 been identified that majority of examined developments are related to rolling-stock maintenanceItand
developments of RAS in railway maintenance domain have been summarized and analysed. has
121 been identified
inspection, which that majority
is about 56%.ofAbout
examined developments are
28% developments are related
relatedto torail-track
rolling-stock maintenance
maintenance and
122 inspection tasks, followed by power transmission maintenance, which is around 5%. maintenance
and inspection, which is about 56%. About 28% developments are related to rail-track RAS in the
123 and inspection
railway bridge tasks, followed
and tunnel by power accounted
maintenance transmission 4%maintenance,
and 3% of total which is around 5%.while
developments, RAS inotherthe
124 railway bridge and tunnel maintenance accounted 4% and 3% of total developments,
maintenance applications, such as automated condition monitoring of grade crossings and cleaning of while other
125 maintenance
stations applications,
and platforms such asfor
accounted automated
4% of the condition monitoring of grade
total developments—see Figurecrossings
3. Further,and cleaning
it has been
126 found that the majority of RAS implemented are limited to inspection and monitoring tasks, which has
of stations and platforms accounted for 4% of the total developments—see Figure 3. Further, it are
127 been found
about 62%. Thethatremaining
the majority 38% ofofRAS
RASimplemented
perform physical are limited to inspection Figure
manipulation—see and monitoring tasks,
4 for a detailed
128 which are about
breakdown by area 62%. The remaining 38% of RAS perform physical manipulation—see Figure 4 for a
of application.
129 detailed breakdown by area of application.

130
Figure 3. Development of Robotic and Autonomous Systems (RAS) in railway maintenance by
131 Figure 3. Development of Robotic and Autonomous Systems (RAS) in railway maintenance by the
the application.
132 application.
 Computers2019,
Computers
Computers 2019,8,
2019, 8,8,x56
xFOR
FORPEER
PEERREVIEW
REVIEW 555ofof
of1616
16

133
133
134
134 Figure 4.
Figure4.
Figure Segregation
4.Segregation of
Segregationof RAS
ofRAS developments
RASdevelopments in
developmentsin railway
inrailway maintenance
railwaymaintenance by
maintenanceby nature
bynature of
natureof task.
oftask.
task.

135 3.1. RAS

3.1.RAS
RASin in Rolling-Stock
inRolling-Stock Maintenance
Rolling-StockMaintenance
Maintenance
135 3.1.
136 As summarized
Assummarized in Figure
summarizedininFigure 5, 36.8%
Figure RASRASdeveloped for rolling-stock maintenance intends intends
to execute
136 As 5,5,36.8%
36.8% developed
RAS developed forrolling-stock
for rolling-stock maintenance
maintenance intends toto
137 physical jobs,
executephysical while
physical the while
jobs, other 63.2%
theother does63.2%
other not perform
doesnot any
not physicalany
perform manipulation. Further, RAS,Further,
physicalmanipulation.
manipulation. which
137 execute
are under study
jobs,
or
while the
fully developed
63.2% does
for rolling-stock
perform
inspection
any physical
and monitoring accounts
Further,
for about
138
138 RAS,which
RAS, whichare areunder
understudy
studyororfully
fullydeveloped
developedfor forrolling-stock
rolling-stockinspection
inspectionand andmonitoring
monitoringaccounts
accounts
139 60% of all RAS
forabout
about 60%of developed
ofall
allRAS for rolling-stock
RASdeveloped
developed maintenancemaintenance
forrolling-stock
rolling-stock tasks, followed by systems
tasks, followedfor byrolling-stock
systemsfor for
139 for
cleaning,
60%
which is 16%. RAS applications
for
in rolling-stock
maintenance tasks,
fluid servicing
followed
tasks account
by systems
for 11% of all
140
140 rolling-stock cleaning, which is 16%. RAS applications in rolling-stock fluid
rolling-stock cleaning, which is 16%. RAS applications in rolling-stock fluid servicing tasks account servicing tasks account
141 rolling-stock
for11%
11%ofofall related
all RAS developments,
rolling-stock relatedRAS RASand the remaining
developments, and13%theofremaining
the applications
remaining 13%ofofconsists of rolling
theapplications
applications
141 for
stock rebuilding
rolling-stock related developments, and the 13% the
142
142 consists
consists rollingand
ofofrolling stock
other
stock applications,
rebuilding
rebuilding andother
and
suchapplications,
other as augmented
applications, suchsuchand virtual
asaugmented
as
reality,
augmented and
and
tovirtual
assist and train
reality,
virtual reality, toto
143 maintenance
assist and workers—see
train maintenance Figure 5.
workers—see Figure 5.
143 assist and train maintenance workers—see Figure 5.

144
144
145 Figure 5.
Figure5. Breakdown
5.Breakdown of
Breakdownof RAS
ofRAS in
RASin rolling
inrolling stock
rollingstock maintenance
stockmaintenance by
maintenanceby area
byarea of
areaof application.
ofapplication.
application.
145 Figure
3.1.1. RAS in Rolling-Stock Cleaning
146
146 3.1.1.RAS
3.1.1. RASininRolling-stock
Rolling-stockcleaning
cleaning
Hiroshi Yaguchi [28] has successfully introduced two robots to clean rolling-stock in East Japan
147
147 HiroshiYaguchi
Hiroshi
Railway
Yaguchi [28]has
company, one[28]
has successfullyintroduced
to cleansuccessfully
introduced tworobots
commuter rail car floorstwo
robotstotoclean
and another
clean rolling-stockininEast
to cleanrolling-stock
EastJapan
the bonnet of Shinkansen,
Japan
148
148 Railway
Railway
high-speed
company,
company,
Japaneseone
one to clean
to clean
trains.
commuter
commuter
A mobile
rail car
raildeveloped
robot is
floors
car floorstoand and another
cleananother
to clean
to cleanrail
the commuter
the
thecarbonnet
bonnet
floors of
of
in
149
149 Shinkansen,
Shinkansen, high-speed
high-speed
conjunction with
Japanese
ToshibaJapanese
trains.
Co. Ltd.trains.
A mobile
A mobile robot
The developed
robot is
robot weighsdeveloped
is developed to clean
to clean the
approximately
the commuter
93 commuter
rail car
rail car
kg, which consists
150
150 floors
floors in conjunction
in conjunction with Toshiba Co. Ltd. The developed robot weighs approximately 93 kg, which
of ultrasonic sensorswithandToshiba
encoders Co.forLtd. The developed
navigation, robot weighs
and optical and touchapproximately
sensor for 93 kg, which
safety. This
151
151 consists
consists of
of ultrasonic
ultrasonic sensors
sensors and
and encoders
encoders for
for navigation,
navigation, and
and optical
optical and
and touch
touch sensor
sensor for
for safety.
safety.
robot can be operated in both automatic and manual modes, and its capabilities include sweeping
152
152 This robot
This robot can
can bebe operated
operated inin both
both automatic
automatic and and manual
manual modes,modes, and and itsits capabilities
capabilities include
include
the floor, collecting dirt, sprinkling water, mopping the floor, pulling dirty water, and polishing the
153
153 sweeping
sweeping the floor, collecting dirt, sprinkling water, mopping the floor, pulling dirty water, and
floor. Thethe floor,
robot collecting
takes about 15dirt,
min.sprinkling
to clean awater,
single mopping the floor, pulling
car. The navigation of robotdirty water,
is done and
by dead
154
154 polishing
polishing the floor.
the floor. TheThe robot takes about 15 min. to clean a single car. The navigation of robot is
reckoning—tracking therobot
desiredtakes
pathabout 15 position
and its min. to clean a single
by encoder car.while
data, The navigation
any positionalof robot
error isis
155
155 doneby
done bydead
deadreckoning—tracking
reckoning—trackingthe thedesired
desiredpath
pathand andits itsposition
positionby byencoder
encoderdata, data,while
whileany any
Computers 2019, 8, 56 6 of 16

corrected using distance sensors [28]. The second robot is developed to clean the Shinkansen train
bonnet with the collaboration of Uchida Oil Hydraulics Mfg Co. Ltd. and ShinMaywa Industries Ltd.
Previously, the cleaning process of Shinkansen bonnet was carried out by six maintenance workers.
The design challenges of the robotic system included demand for a large work envelope, compact rest
position, and ability to move along and lateral directions to the train. Therefore, researchers have
designed an eight-axis articulated robot mounted on a mobile platform that was capable of moving
along the full-length of a Shinkansen train and across three service bays in the lateral direction. The
main control method of the robot was point-to-point teaching and playback. Hiroshi has concluded
the main challenges in both RAS applications to be cost and ability to sense environment/target
effectively [28].
A self-traction-model to clean urban mass transit vehicles was proposed by Xu et al. The
washing apparatus is mounted on a mobile vehicle platform, which is driven by a variable speed and
voltage inverter. The motion and path planning of the manipulator is achieved by the neutral cell
self-adaptive and Cross-Coupling Control algorithm. Further, a novel variable water control algorithm
is incorporated into the system to make it energy and water efficient. The researchers claim that
the system has gone through years of testing in various depots and was proven to work efficiently
and robustly [29]. Further, there are some conceptual designs and ongoing research to introduce
RAS in rolling-stock cleaning tasks. For example, Tomiyama et al. [30] have presented a systematic
analysis of the train cab front cleaning task and generated subclasses that will enable researchers to
efficiently and effectively find solutions to each sub-task. Further, this study presents several conceptual
designs of cab front cleaning robots and two illustrations of robotic arms are provided. In contrast to
Hiroshi Yaguchi’s bonnet cleaning robot, Tomiyama et al. have divided the work envelope into two
along the vertical symmetric line of the train and positioned two identical robots in each side [28,30].
Furthermore, Moura and Erden [18] presented a theoretical control and path planning methodology
for a train cab front cleaning robot by exploiting the operational space formulation and simultaneous
force and position control introduced by Khatib [31,32].

3.1.2. RAS in Rolling-Stock Fluid Servicing

Thus far, the authors were unable to locate a RAS, which is successfully implemented in a
rolling-stock fluid changing task. However, there are several ongoing studies to introduce RAS in
rolling-stock fluid changing tasks. Atherton et al [15] have conducted a technical feasibility study in
introducing RAS to passenger train fluids, such as fresh water, coolant, screen wash, fuel, effluent, and
wheel sand servicing tasks. One of the challenges identified in this research is the different sizes and the
positioning of fluid ports in fleets. Through this study, it has been concluded that the implementation
of RAS in passenger train fluid servicing task is viable. Further, two preliminary conceptual designs,
which consist of a Cartesian robot and an articulated robot, are proposed.
Further, Farnsworth and Tomiyama [33] looked into automated transmission fluid changing task
of rolling stock with an industrial engineering aspect. They have introduced a novel methodology to
capture, classify, and design automated platforms for industrial maintenance based on a case study
that was conducted on Bombardier Class 222 diesel train transmission fluid changing process. In
contrast, work that was carried out by Vithanage et al. [14] emphasizes the technical aspect of the
gear fluid changing tasks as compared to other high-level investigations conducted. Through this
research, they have successfully adopted Aggregated Channel Feature, an advanced machine learning
algorithm to detect the drain and filler plugs of Siemens Class 380 rolling-stock main gearbox and
curved coupler. Further, they have fused multiple sensors and introduced an Artificial Neural Network,
which consists Levenberg–Marquardt learning algorithm to localize relatively smaller components in
partially organised worlds, such as railway maintenance depots with ±1 mm accuracy [34].
Computers 2019, 8, 56 7 of 16

3.1.3. RAS in Rolling-Stock Inspection

Rolling-stock inspection is expensive, labor-intensive, and critical for safe operation. Therefore,
certain inspection tasks have been automated over the past several decades. For example, the
implementation of wayside detectors, cameras mounted to the rolling-stock undercarriage, and
automated inspection tasks in maintenance depots [35]. In most cases, rolling-stock wheels and
bearings are responsible for derailments [36]. Therefore, a number of automated wayside detectors
are designed and developed to overcome this issue. In Conrail, the derailments due to defects in the
wheels were reduced almost 50% by introducing wayside detectors [36]. Papaelias et al. have proposed
a rolling-stock wheel and axle bearing monitoring system that is based on high-frequency acoustic
emission and vibration analysis mounted to a train. The study has found that it is viable to detect
and classify axle bearing defects by integrating high-frequency acoustic emission data with vibration
data based on their severity [4]. However, these wayside monitoring systems need to be robust and
withstand harsh weather conditions. Therefore, Asplund et al. have tested an automatic laser-based
wheel profile monitoring system (WPMS) for accuracy and performance in extreme weather conditions.
The study shows that laser-based WPMS indicates a deviation of less than 0.2 mm for the flange height,
0.3 mm for the flange thickness, and 0.32 mm for the flange slope as compared to measurements
collected through handheld MiniProf measurement equipment [37].
Hart et al. have proposed a multi-spectral rolling-stock undercarriage inspection technique that
combines visible and infrared images, which allow for recording both physical and thermal conditions,
and the correlation between two. The scope of this study includes the monitoring and inspection
of disc brake condition, bearing performance, and detection of any anomalies in electrical systems,
such as induction motor and air-condition unit [38]. A similar thermal imagery approach is used by
Deilamsalehy et al. to detect sliding wheels and hot bearings. The proposed methodology consists of a
wheel and bearing detection module, hot bearing detection module using thresholding, and a Support
Vector Machine (SVM) classifier, which uses a Histogram Oriented Gradient (HOG) of thermal images
as input to detect flat wheels. The developed system is capable of detecting simulated and real-world
defective wheels, with 98% accuracy [39]. Furthermore, a study has been conducted to determine the
feasibility of inspecting rail underframe components of freight cars, such as center sill, side-sills, and
cross-bearers. The study has proposed both a hardware and software to collect the required visual data
and detect and assess rail undercarriage components for any deformation or buckling and presence of
breaks or cracks. The study concludes that presented image acquisition and machine vision algorithms
demonstrated the feasibility of automating the inspection of structural components [40].
Liu et al. have proposed an automated status inspection of fastening bolts on freight trains while
using machine vision. The developed algorithm delivered the best performance by combining gradient
orientation co-occurrence matrix and SVM classifier. The system is able to detect the existence or
nonexistence of a bolt in a complex environment with 99.96% accuracy and it is expected to evaluate
the performance of the system in adverse weather conditions [41]. In addition, Li et al. have proposed
another automatic bolt defect recognition system—which focuses on freight train wheels. Similar to
the method that was proposed by Liu et al., they have utilized the SVM classifier to distinguish the
presence or absence of a bolt. In contrast, Li et al. have used a binary pattern descriptor as input to the
classifier and their system consists of a self-updating scheme, which automatically captures various
real-life scenarios. The system, as presented by Li et al., has demonstrated 100% bolt defect detection
rate with 8.1% false alarm rate [42].
Additionally, Kim and Kim have proposed an automatic system that accurately measures brake
shoe thickness with an accuracy of 0.654 mm. They have successfully exploited the shadow regions
between the brake shoe and wheels that look similar in all images to localize the brake shoes by
modeling the boundaries as inverse polynomials [43]. Furthermore, Zhou et al. have presented
a real-time automated visual inspection system of angle cocks, which is a critical component of a
rolling-stock brake system. The developed system incorporates a gradient encoding histogram and
SVM in the processing module. It has been claimed that the system achieves 99.8% accuracy in fault
Computers 2019, 8, 56 8 of 16

detection [44]. In addition, the Bombardier team at Midrand, South Africa has commissioned an
Automatic Vehicle Inspection System (AVS), together with expertise in MRX technologies and IBM,
which is proven to be extremely successful. The system is capable of monitoring multiple aspects, such
as brake pad and disc conditions, axle temperature, wheel profile, wheel damage, pantograph wear,
and any abnormalities on train exterior profile while using cutting-edge three-dimensional (3D) image
acquisition and processing tools [45].
Edwards at al. and Liu et al. have presented online visual inspection systems that capable
inspecting rail car safety appliances and bogie block keys of freight trains respective. Both systems
utilize machine vision, advanced image processing algorithms, and state of the art machine learning
algorithms to identify the defects [46,47]. Besides, Vithanage et al. have presented an image processing
system to identify and localize the connector pins of Scharfenberg type automatic couplers mounted
on Siemens Class 380 trains. The system claims to have high detection rate and robustness to
environmental noises. The proposed system expects containing an off the shelf industrial robot to
execute the inspection process [14]. This is system was further developed to have a unified 3D vision
and a force-position controller. The prototyped work cell is validated while using an industrial grade
6-DoF articulated robot. It has been estimated to improve the standard operating time by approximately
34% [48].

3.2. RAS in Rolling-Stock Rebuilding

Kent and Ward have carried out some early development work to introduce robotics into the
additive repair of rolling-stock wheels, which is supported by the Rail Safety and Standards Board
(RSSB) via Rail Research UK Association (RRUKA) [10]. However, as an intermediate output, they
have proposed a rolling-stock wheelset inspection robotic cell that consists of a 6 DoF articulated
Kuka robot mounted on a rail, ToF sensor, RGB sensor, and a train wheel rotating mechanism. In
addition, the mining company Vale, Brazil has invested on a robotic rail car rebuilding cell. According
to the manager, operational improvements at Vale, maintenance cycle time was reduced from 48
hours per car to 11 hours per car on average. Further, they have managed to remove technicians from
labour intensive and dangerous work, and assign them to safer and more value-added tasks [49].
A commercial rolling-stock wheel set maintenance shop at “SIMMONS Machine Tool Corporation”
utilizes reprogrammable robots and overhead gantry system to automate the movement of the wheel
sets and their components between various workstations and machines. The system claims to decrease
human contact with wheel set components and machines, improve personal health and safety, reduce
opportunities for error, and eliminate the requirement of forklift operation on the shop floor [50].

3.3. RAS in Power Transmission Line Maintenance

A power transmission wire inspection system was developed for Netherlands railway company,
which perceives the thickness of wires and detects any appearance of holes that are caused by local
sparks [51]. In this system, the under surface of the overhead wire is illuminated with a laser beam and
reflected radiation is then detected by a fast CCD detector. The processing algorithm determines the
thickness of wire and any of occurrence of holes. According to Smorenburg and Valkenburg, the system
is capable of mounting on a train, which travels 90km/h and inspects four wires simultaneously for
every centimetre. Hence, the system is able to efficiently execute inspection process without disrupting
normal services [51]. In addition, Sawada et al. have developed a fully autonomous robot to inspect
power transmission lines—which is capable of negotiating obstacles, such as towers, ground wires,
etc. The robot consists of curved arm—which helps to negotiate obstacles, a vehicle assembly that
carries the inspection unit, and a balancer with a built-in controller. The robot measures eddy current
and the processing algorithm determines the condition of the power transmission wire based on the
input data [52]. Besides, Peungsungwal et al. have developed an autonomous power transmission line
inspection robots to examine the power cables and insulators [53]. The robot uses image-processing
techniques, such as edge detection and pattern recognition algorithms to detect insulators. The most
Computers 2019, 8, 56 9 of 16

interesting feature of this robot is that it does not depend on a battery pack or any other sources of fuel,
but harvest the required energy from power line through induction [53].

3.4. RAS in Track Maintenance

Railway track maintenance plays a major role in the railway industry. Trivedi et al. have proposed
a fully autonomous robotic system to detect and remove sleeper bolts, feed new fastener, and assemble
them in Shinkansen high-speed train lines. Instead of creating one massive robot to perform all
tasks, researchers have taken a different approach and created a multi-robotic system, where the
individual robot will focus on a simplified task. According to Trivedi et al., the system is technically and
economically viable [54]. Further, Rowshandel et al have proposed an integrated robotic system for the
detection and characterization of rolling contact fatigue cracks in rails. The suggested system consists
of a mechanized trolley, a robot, a commercially available alternating current field measurement
(ACFM) system, and a laser distance sensor. The trolley, which contains the articulated robot, will
travel at a controlled speed scanning for surface cracks on the rail and the robot will take over the
scanning process while using the ACFM system once a crack is detected. Further, it has been concluded
that the system manages to autonomously detect and characterize surface-breaking defects with high
precision [55].
Furthermore, researchers have successfully exploited the recent developments of machine vision
and classification algorithms to develop RAS in rail track maintenance tasks, especially for inspection
and monitoring [56,57]. Santur et al. have proposed a robust rail condition monitoring methodology
while using a laser scanner mounted on a moving rail car, which detects fractures, scorings, and
excessive wear of the track. Random Forest classification learning algorithm is used in their system by
considering the Eigenvalues of rail profiles as input. According to the authors, the system demonstrates
98% accuracy [58]. A similar track geometry diagnostic system has been developed by Madejski, which
alerts the weak locations of tracks and helps to save about 90%–95% labour consumption [59].
In addition, there are few dozens of RAS that have been developed to detect the missing or worn
out track components, such as bolts, clips, ties, tie plates, anchors, and turnout components [60–62].
For example, an autonomous clip detection and classification system was introduced by Gibert et al.,
where HOG of target images was analyzed to detect clips and classify their condition as good, missing,
or broken by taking advantage of Linear Support Vector Machine (LSVM) learning algorithm. The
described system is capable of achieving a probability of detection of 98% and a false alarm rate of
1.23% [63]. Another application is developed to detect missing railway clips and recently replaces clips
that are different in color by employing basic image processing techniques, such as Gaussian filtering,
Canny edge detection, suppressing smaller edges from a binary image, analyzing edge pixels density
of binary images, and histogram of colors. According to Singh et al., the system is accurate up to 95.3%,
86.5%, and 84.7% in finding clips, blue clips, and missing clips, respectively [64]. In addition, Li et al
present a real-time automatic vision-based rail inspection system, which performs inspections at 16
km/h with a frame rate of 20 fps to detect and inspect tie plate, tie, and anchor. The proposed system
that is composed with an image and video analysis methodology, which is combined with multiple
cameras, Global Position System (GPS) and a distance measurement instrument. The system consists
of an acquisition module, rail component detection, optimization module, and an exception detection
module to detect damaged or missing components. Further, the systems acquired following detection
results, Tie plate: 100% recall rate and 99.3% precision, Tie: 82.3% recall rate and 88.2% precision, and
Anchor: more than 96% recall rate and precision [65].
A higher technology readiness level robotic solution for rail track inspection is developed
by “Autoscan”, which uses electromagnetic acoustic transducers (EMATs) that are mounted to a
self-powered autonomous inspection cart with a robotic arm. The system is at TRL 6 and it has the
ability to reduce track inspection costs by at least 15%. It is expected to record different defects with an
accuracy of +/- 0.5 mm [66]. In addition, a commercially available automated rail inspection robotic
platform has been introduced by “RailPod”. The platform is composed of a vehicle that can be remotely
Computers 2019, 8, 56 10 of 16

controlled or drive autonomously and it contains the ability to configure based on individual customer
need [67].

3.5. RAS in Bridge Maintenance

Railway bridge maintenance is labor-intensive and, in most circumstances, technicians need to
execute the bulk of tasks in dull, dirty and dangerous environments. Therefore, to relieve human
workers from such labour intensive and dangerous tasks, Liu et al. have introduced a robotic system
for stripping paint and rust from steel bridges. The proposed system consists of a 6-DOF industrial
robot, a moving platform, a sensor package that is equipped with a laser range finder, cameras, and a
capacitive sensor network, and a high-performance computer [68]. In addition, Wang and Kawamura
have designed and developed a semi-autonomous magnetic climbing robot for steel bridge inspection,
which can be freely moved without disrupting the traffic. The magnetic wheels of this design not only
provide necessary adhesion, but the required magnetic field for inspection. The system is yet to be
completed and tested on real bridge inspection task [69]. Furthermore, Oh et al. have prototyped a
bridge inspection robot, which consists of a specially designed car, a guide rail, and an inspection robot.
The proposed system is capable of fully autonomous operation or teleoperation by distant user, which
has been successfully tested in laboratory conditions. It is expected to develop the robot to operate in
real environments [70].

3.6. RAS in Tunnel Maintenance

The majority of railway tunnels are inspected by human technicians and that suffers from human
errors, costs, and high inspection time [71]. Further, atmospheric conditions within tunnels are
characterized as dusty, humid, and dark, which acts as hazardous and hostile environments for
human workers [72]. IRIS Hyrail by Penetradar is a commercially available semi-autonomous tunnel
inspection system, which consists of a Ground-penetrating radar (GPR). The inspection module is
fixed to a motorized boom, which is mounted on a vehicle that can be driven on the rail tracks. The
manufacturer provides specialized software to manage, collect, and display GPR data [73]. In addition,
Aoshima et al. have proposed an automatic tunnel inner wall deterioration monitoring system. The
suggested scheme consists of sensors that are fused to regular train (velocity more than 100 km/h)
which is then used capture data on daily basis. The system is capable of scheduling the necessary
maintenance upon the detection of any deterioration being discovered by analysing the relationship
between the train velocity and the cross-correlation coefficient of the interpolated crack images [71].

3.7. RAS in other Railway Maintenance Tasks

RAS in other railway maintenance tasks include automated cleaning robots in train stations,
monitoring of railway infrastructure using drones, monitoring and inspection of grade crossing at
etc. Hiroshi has introduced two mobile cleaning robots, to clean stations in the East Japan Railway
Company [28]. One robot is designed to sweep the and collect the dust that accumulated in the
corners of station floor and another robot is designed to scrub the station floor. The sweeper type
robot has three operational modes, namely ‘magnetic guide method’, where the robot follows the
path of a magnetic tape, ‘direct teaching method’, where the robot memorizes the path taught by an
operator, and a ‘wall surface copy system’, where the robot follows the wall while using ultrasonic
distance sensors.
The second robot, by Hiroshi, is capable of sprinkling water, scrubbing the floor, and removing
consumed water by suction. The robot is fixed with a self-propelled system that includes a gyroscope
in order to cope with wider station concourse. This robot consists of two operating modes, ‘area set up
mode’, where the robot cleans the input length and width of the area and ‘route map mode’ where the
robot cleans according to a built-in map. The area clean mode has the obstacle avoidance capability,
while the built-in map mode does not. Furthermore, it is possible to upload up to nine built-in maps.
Both of these robots consist of safety features, such as optical sensors to detect the presence of humans,
Computers 2019, 8, 56 11 of 16

melody chime, alarm lamp, emergency stop switches, and soft bumpers. A similar commercially
available cleaning robot for various industrial applications, including railway platforms, is available
at “Robotics Inventions” under original equipment manufacturer license. The system claims to have
military grade autonomy and mapping capabilities, automated recognition of dirt type to be cleaned,
and auto-service station with recycled water tank [74].
Besides, Flammini et al. have evaluated the drone capabilities in a railway-monitoring framework,
which include structural health monitoring, such as bridge, tunnel, and other infrastructure inspection,
track monitoring, and inspection to detect any faults or disturbances. Authors have presented several
potential control and communication frameworks for the drones and claim that drones could represent
a promising technology to efficiently monitor railway and mass transit infrastructures [75]. In addition,
Ranganathan and Olson have proposed an automated grade crossing condition monitoring system.
The required sensors are mounted to a vehicle and the developed algorithm is capable of processing
data on the go. The presented system is able to identify a grade crossing with 90% accuracy while
using SVM and identify any dangerous traffic flow lines/regions based on the surface profile of the
grade crossing [76,77].
A robotized, semi-automated platform is introduced in [78,79] to collect different safety parameter
measurements that are linked to railway track circuits (TC). A system architecture, together with a
virtual instrumentation and mathematical model, is presented. It has been concluded that the proposed
system can become a viable solution for TC maintenance. Authors claim that this system would be an
efficient solution for current manual maintenance practice and aim to improve the operator safety and
reduce human errors.
Railway maintenance as a whole is typically a large and complex system. Therefore, these tasks
needs to be scheduled in an effective way to improve safety, reduce downtimes, eliminate corrective
maintenance jobs, optimize asset/equipment life cycle, and reduce costs. One approach to overcome
this challenge is to exploit a data rich environment of railway and introduce autonomous scheduling
for maintenance jobs. In [80], an automated task scheduling scheme is introducing by fusing data on
railway’s condition, planning, and costs. Here, a proof of concept is developed to validate the system
and test algorithm functionality. Its authors conclude that the demonstrator found to be viable fit for
the purpose. It is expected to conduct further research and evaluate commercial exploitation of the
system. A similar system is introduced in [81]. Here, a genetic algorithm is used to optimize railway
track maintenance and renewal job scheduling. The system was demonstrated to perform well and
illustrated the ability to instead use corrective maintenance activities.

4. Discussion
The study reveals a strong case for the necessity of RAS in railway maintenance to improve
the safety of passengers and railway workers, quality and reliability of the service, and to mitigate
unpredicted expenses through preventive maintenance. It has been revealed that the majority of
RAS developments focus on rolling-stock and rail track maintenance tasks, which are justifiable, as
said constituents play a major role in the railway industry. Further, it has been realized that the
majority of RAS developments in the subject area focus on inspection and monitoring, which are
maintain the advantage of computer vision and supervised machine learning algorithms. Therefore, it
would be advantageous if more research were conducted on applied robotics to elevate the mechanical
engineering aspect of such systems. For instance, introduce collaborative robots to support maintenance
technicians in railway depots [82], develop a robotic system to conduct scheduled maintenance tasks,
such as front end inspection and service of rolling-stocks or to service rolling-stock brake system,
equip railway workers who handle heavy tasks with load carrying exoskeletons to minimize muscular
fatigues [83], or adopt humanoid robots from service robotic sector to handle complex railway
maintenance tasks [77,84,85].
On the other hand, the majority of inspection tasks tend to exploit SVM, which is a supervised
machine algorithm to detect the presence of components or to classify the condition of a component
Computers 2019, 8, 56 12 of 16

as good or damaged. For example, there are several applications for detecting missing bolts and
screws, but it emerges that there is not enough research to detect a loose screw or bolt. Further, more
examinations are required to exploit different classification algorithms other than SVM in the railway
maintenance field. In addition, the detection of smaller railway components in unstructured and
dynamic environments is non-trivial. For instance, the detection and localization of connectors of the
automatic train coupler electric head pins or detection of wire-locks used to secure drain/filler plugs in
certain rolling-stock gearboxes, such as Siemens Class 380 [14].
Furthermore, it is required to make robotic systems as flexible as possible while using off the
shelf robots than introducing fully customized designs that yield higher costs, deficit in flexibility, and
longer development periods [79]. Flexible and easily reconfigurable RAS developments will provide
the ability to deploy the RAS in multiple tasks to attain benefits, such as shorter financial recovery
periods and space saving. Further, such systems are viable through the proper design of layout, path
planning, and development of intelligent and reconfigurable end of arm tooling. For instance, deploy
a robot mounted on a rail or a scissor lift to clean train cab front and inspect automatic train coupler or
utilize a robot that is mounted on a moveable platform to perform both transmission fluid and brake
service tasks.

5. Conclusions
Not to overlook the rapid progress made by RAS in railway maintenance sector, there are still
ample areas to be investigated. Authors of the paper believe that, in the near future there will be
collaborative robots working together with human workers in railway maintenance. Lastly, the
introduction of RAS for railway maintenance cannot be a viewed as a sole responsibility of academic
researchers, operations, maintenance, or industrial engineers. For instance, rolling-stock designers
and manufacturers should consider promising areas to introduce RAS in rolling-stock maintenance
and incorporate strategies to their future designs—which will enable the realization autonomous
maintenance systems. In other words, designed and manufactured for automated maintenance.

Author Contributions: All three authors have contributed equally for this article.
Funding: This research received no external funding.
Acknowledgments: This work is supported by Glasgow Caledonian University. Also, authors would like to thank
Gary McLachlan of Abellio ScotRail, Glasgow, Roger Harris of Cyber-Weld Ltd, Tony Rodden and Alexandros
Gkanatsios at Advanced Forming Research Centre, UK.
Conflicts of Interest: The authors declare no conflict of interest

References
1. Márquez, A.C. The Maintenance Management Framework: Models and Methods for Complex Systems Maintenance;
Springer Science & Business Media: Berlin, Germany, 2007.
2. Reason, J. Cognitive Engineering in Aviation Domain; Lawrence Erlbaum Associates: Mahwah, NJ, USA, 2000;
Volume 72, pp. 102–107.
3. Holmgren, M. Maintenance-related losses at the Swedish Rail. J. Qual. Maint. Eng. 2005, 11, 5–18. [CrossRef]
4. Papaelias, M.; Amini, A.; Huang, Z.; Vallely, P.; Dias, D.C.; Kerkyras, S. Online condition monitoring of
rolling stock wheels and axle bearings. J. Rail Rapid Transit 2016, 230, 709–723. [CrossRef]
5. Ciocoiu, L.; Siemieniuch, C.E.; Hubbard, E.-M. From preventative to predictive maintenance:
The organisational challenge. J. Rail Rapid Transit 2017, 231, 1174–1185. [CrossRef]
6. Cores, F.; Caceres, N.; Benitez, F.; Escriba, S.; Jimenez-Redondo, N. A logical framework and integrated
architecture for the rail maintenance automation. In Proceedings of the European Transport Conference 2013
Association for European Transport (AET), Frankfurt, Germany, 30 September–2 October 2013.
7. ARUP Group. Rail Value for Money Study: Rolling Stock Whole Life Costs; Department for Transport and the
Office of Rail Regulation: London, UK, 2011.
8. National Audit Office (NAO). A Short Guide to Network Rail; National Audit Office: London, UK, 2015.
Computers 2019, 8, 56 13 of 16

9. Jimenez-Redondo, N.; Bosso, N.; Zeni, L.; Minardo, A.; Schubert, F.; Heinicke, F.; Simroth, A. Automated
and Cost Effective Maintenance for Railway (ACEM–Rail). Procedia Soc. Behav. Sci. 2012, 48, 1058–1067.
[CrossRef]
10. Stephen Kent, M.W. Feasibility of the Use of Autonomous Robotic Systems for Wheelset Reworking. Level
of Thesis, University of Birmingham, Birmingha, UK, 2017.
11. Preston, J.; Wall, G.; Batley, R.; Ibáñez, J.; Shires, J. Impact of delays on passenger train services: Evidence
from Great Britain. Transp. Res. Rec. 2009, 2117, 14–23. [CrossRef]
12. Eversholt Rail Group RDG Angel Trains Porterbrook. Long Term Passenger Rolling Stock Strategy for the Rail
Industry [Internet]. Rail Delivery Group. 2017. Report No.: 5. Available online: https://www.midlandsrail.
co.uk/wp-content/uploads/2017/04/Rolling-Stock-Strategy-5th-ed.-2017-published-by-RDG.pdf (accessed
on 8 February 2019).
13. Brage-Ardao, R.; Graham, D.J.; Anderson, R.J. Determinants of rolling stock maintenance cost in metros.
J. Rail Rapid Transit 2016, 230, 1487–1495. [CrossRef]
14. Vithanage, R.K.W.; Harrison, C.S.; DeSilva, A.K.M. A Study on Automating Rolling-stock Maintenance in the
Rail Industry using Robotics. In Proceedings of the 14th International Conference on Informatics in Control,
Automation and Robotics, Madrid, Spain, 26–28 July 2017; Volume 2, pp. 278–283.
15. Atherton, M.; Hill, S.; Harrison, D.; Ajovalasit, M. Economic and technical feasibility of a robotic autonomous
system for train fluid servicing. J. Rail Rapid Transit 2019, 0954409719830520. [CrossRef]
16. Sanne, J.M. Framing risks in a safety-critical and hazardous job: Risk-taking as responsibility in railway
maintenance. J. Risk Res. 2008, 11, 645–658. [CrossRef]
17. Feng, H.; Jiang, Z.; Xie, F.; Yang, P.; Shi, J.; Chen, L. Automatic fastener classification and defect detection in
vision-based railway inspection systems. IEEE Trans. Instrum. Meas. 2014, 63, 877–888. [CrossRef]
18. Moura, J.; Erden, M.S. Formulation of a control and path planning approach for a cab front cleaning robot.
Procedia Cirp. 2017, 59, 67–71. [CrossRef]
19. Itoh, K.; Andersen, H.B.; Seki, M. Track maintenance train operators’ attitudes to job, organisation and
management, and their correlation with accident/incident rate. Cogn. Technol. Work 2004, 6, 63–78. [CrossRef]
20. Singh, S.; Kumar, R.; Barabadi, A.; Kumar, S. Human error quantification of railway maintenance tasks of
disc brake unit. In Proceedings of the World Congress on Engineering 2014, London, UK, 2–4 July 2014.
21. McNulty, R. Realising the Potential of GB Rail: Final Independent Report of the Rail Value for Money Study;
Department for Transport and the Office of Rail Regulation: London, UK, 2011.
22. Technical Strategy Leadership Group (TSLG). The Future Railway: The Industry’s Rail Technical Strategy
2012 Supporting Railway Business; Technical Strategy Leadership Group & Rail Safety and Standards Board:
London, UK, 2012.
23. Taylor, M.; Westwood, M.; Peters, A. A modular approach to automation of condition monitoring and repair
for rail. In Proceedings of the 7th IET Conference on Railway Condition Monitoring (RCM), Birmingham,
UK, 27–28 September 2016; pp. 1–6.
24. Rail Research UK Association (RRUKA). Applications of Robotics and Autonomous Systems to Rolling Stock
Maintenance; Rail Research UK Association: London, UK, 2015.
25. Shioya, A.; Sasama, H. Inspection robots for railways in Japan. Adv. Robot. 1988, 3, 221–227. [CrossRef]
26. Martland, C.D. Analysis of the potential impacts of automation and robotics on locomotive rebuilding. IEEE
Trans. Eng. Manag. 1987, EM-34, 92–100. [CrossRef]
27. Wiercienski, W.; Leek, A. Feasibility of robotic cleaning of the undersides of Toronto subway cars. In
Proceedings of the IEEE 39th Vehicular Technology Conference 1989, San Francisco, CA, USA, 1–3 May 1989;
pp. 884–887.
28. Yaguchi, H. Robot introduction to cleaning work in the East Japan Railway Company. Adv. Robot. 1995, 10,
403–414. [CrossRef]
29. Xu, H.; Xue, K.; Yu, S.; Gao, X.; Qing, C.; Wei, R. Control for an Innovative Robotics Platform of Rinsing
System for Vehicles of Urban Mass Transit. In Proceedings of the 2007 4th IEEE International Conference on
Mechatronics, Changchun, China, 8–10 May 2007; pp. 1–6.
30. Tomiyama, T.; García, L.R.; Kršlin, A.; Taykaldiranian, G. Systems and conceptual design of a train cab front
cleaning robot. Procedia Cirp 2017, 59, 61–66. [CrossRef]
Computers 2019, 8, 56 14 of 16

31. Khatib, O.; Burdick, J. Motion and force control of robot manipulators. Robotics and Automation. In
Proceedings of the 1986 IEEE International Conference on Robotics and Automation, San Francisco, CA,
USA, 7–10 April 1986; pp. 1381–1386.
32. Khatib, O. A unified approach for motion and force control of robot manipulators: The operational space
formulation. IEEE J. Robot. Autom. 1987, 3, 43–53. [CrossRef]
33. Farnsworth, M.; Tomiyama, T. Capturing, classification and concept generation for automated maintenance
tasks. CIRP Ann. Manuf. Technol. 2014, 63, 149–152. [CrossRef]
34. Vithanage, R.K.; Harrison, C.S.; DeSilva, A.K. Enhance 3D Point Cloud Accuracy Through Supervised
Machine Learning for Automated Rolling Stock Maintenance: A Railway Sector Case Study. In Proceedings
of the 2018 International Conference on Computing, Electronics & Communications Engineering (iCCECE),
Southend, UK, 16–17 August 2018; pp. 242–246.
35. Kilian, K.; Kilian, M.; Mazur, V.; Phelan, J. Rethinking reliability engineering using machine vision systems.
J. Rail Rapid Transit 2016, 230, 1006–1014. [CrossRef]
36. Steets, P.G.; Tse, Y.H. Conrail’s integrated automated wayside inspection. In Proceedings of the 1998
ASME/IEEE Joint Railroad Conference, Philadelphia, PA, USA, 16 April 1998; pp. 113–125.
37. Asplund, M.; Gustafsson, P.; Nordmark, T.; Rantatalo, M.; Palo, M.; Famurewa, S.M.; Wandt, K. Reliability
and measurement accuracy of a condition monitoring system in an extreme climate: A case study of automatic
laser scanning of wheel profiles. J. Rail Rapid Transit 2014, 228, 695–704. [CrossRef]
38. Hart, J.; Resendiz, E.; Freid, B.; Sawadisavi, S.; Barkan, C.; Ahuja, N. Machine vision using multi-spectral
imaging for undercarriage inspection of railroad equipment. In Proceedings of the 8th World Congress on
Railway Research, Seoul, Korea, 18 May 2008.
39. Deilamsalehy, H.; Havens, T.C.; Lautala, P.; Medici, E.; Davis, J. An automatic method for detecting sliding
railway wheels and hot bearings using thermal imagery. J. Rail Rapid Transit 2017, 231, 690–700. [CrossRef]
40. Schlake, B.W.; Todorovic, S.; Edwards, J.R.; Hart, J.M.; Ahuja, N.; Barkan, C.P. Machine vision condition
monitoring of heavy-axle load railcar structural underframe components. J. Rail Rapid Transit 2010, 224,
499–511. [CrossRef]
41. Liu, L.; Zhou, F.; He, Y. Automated status inspection of fastening bolts on freight trains using a machine
vision approach. J. Rail Rapid Transit 2016, 230, 1629–1641. [CrossRef]
42. Li, C.; Wei, Z.; Xing, J. Online inspection system for the automatic detection of bolt defects on a freight train.
J. Rail Rapid Transit 2016, 230, 1213–1226. [CrossRef]
43. Kim, H.; Kim, W.-Y. Automated thickness measuring system for brake shoe of rolling stock. In Proceedings
of the 2009 Workshop on Applications of Computer Vision (WACV), Snowbird, UT, USA, 7–8 December
2009; pp. 1–6.
44. Zhou, F.; Zou, R.; Qiu, Y.; Gao, H. Automated visual inspection of angle cocks during train operation. J. Rail
Rapid Transit 2014, 228, 794–806. [CrossRef]
45. Lortie, M.; St-Bruno, Q.; Holmes, E.; UK Burton upon Trent. Automated Inspection Technologies: A New
Paradigm for Preventive Maintenance Programs. In Proceedings of the 2014 Rail Conference American
Public Transportation Association, Houston, TX, USA, 12–15 October 2014.
46. Edwards, J.R.; Hart, J.M.; Todorovic, S.; Barkan, C.P.; Ahuja, N.; Chua, Z.; Kocher, N.; Zeman, J. Development
of machine vision technology for railcar safety appliance inspection. In Proceedings of the International
Heavy Haul Conference Specialist Technical Session-High Tech in Heavy Haul, Kiruna, Sweden, 11–13 June
2007; pp. 745–752.
47. Liu, L.; Zhou, F.; He, Y. Automated Visual Inspection System for Bogie Block Key Under Complex Freight
Train Environment. IEEE Trans. Instrum. Meas. 2016, 65, 2–14. [CrossRef]
48. Vithanage, R.K.; Harrison, C.S.; De Silva, A.K. Autonomous rolling-stock coupler inspection using industrial
robots. Robot. Comput. Integr. Manuf. 2019, 59, 82–91. [CrossRef]
49. Vale. Robot in the Maintenance of Cars [Internet]. 2016. Available online: http://www.vale.com/brasil/EN/
aboutvale/news/Pages/robos-manutencao-vagoes.aspx (accessed on 18 November 2018).
50. Corporation SMT. Automated Material Handling [Internet]. 2015. Available online: https://smtgroup.com/
en/railway-wheel-shop-equipment/automation/automated-material-handling (accessed on 20 Januar 2019).
51. Smorenburg, C.; van Valkenburg, A.L.G. Automatic inspection of railway overhead wires. In Industrial
Inspection; International Society for Optics and Photonics: Bellingham, WA, USA, 1989; pp. 107–117.
Computers 2019, 8, 56 15 of 16

52. Sawada, J.; Kusumoto, K.; Maikawa, Y.; Munakata, T.; Ishikawa, Y. A mobile robot for inspection of power
transmission lines. IEEE Trans. Power Deliv. 1991, 6, 309–315. [CrossRef]
53. Peungsungwal, S.; Pungsiri, B.; Chamnongthai, K.; Okuda, M. Autonomous robot for a power transmission
line inspection. In Proceedings of the 2001 IEEE International Symposium on Circuits and Systems, Sydney,
Australia, 6–9 May 2001; pp. 121–124.
54. Trivedi, M.M.; Ng, K.C.; Lassiter, N.; Capella, R. New generation of multirobot systems. In Proceedings of
the 1998 IEEE International Conference on Systems, Man, and Cybernetics, San Diego, CA, USA, 14 October
1998; pp. 3342–3346.
55. Rowshandel, H.; Nicholson, G.L.; Davis, C.L.; Roberts, C. An integrated robotic system for automatic
detection and characterisation of rolling contact fatigue cracks in rails using an alternating current field
measurement sensor. J. Rail Rapid Transit 2013, 227, 310–321. [CrossRef]
56. Mazzeo, P.L.; Nitti, M.; Stella, E.; Distante, A. Visual recognition of fastening bolts for railroad maintenance.
Pattern Recognit. Lett. 2004, 25, 669–677. [CrossRef]
57. Resendiz, E.; Hart, J.M.; Ahuja, N. Automated visual inspection of railroad tracks. IEEE Trans. Intell. Transp.
Syst. 2013, 14, 751–760. [CrossRef]
58. Santur, Y.; Karaköse, M.; Akin, E. Condition Monitoring Approach Using 3d Modelling Of Railway Tracks
With Laser Cameras. In Proceedings of the International Conference on Advanced Technology & Sciences
(ICAT’16), Konya, Turkey, 1–3 September 2016; pp. 132–135.
59. Madejski, J. Autonomous track geometry diagnostics system. J. Mater. Process. Technol. 2004, 157, 194–202.
[CrossRef]
60. Wang, L.; Zhang, B.; Wu, J.; Xu, H.; Chen, X.; Na, W. Computer vision system for detecting the loss of rail
fastening nuts based on kernel two-dimensional principal component–two-dimensional principal component
analysis and a support vector machine. J. Rail Rapid Transit 2016, 230, 1842–1850. [CrossRef]
61. Zhang, H.; Yang, J.; Tao, W.; Zhao, H. Vision method of inspecting missing fastening components in
high-speed railway. Appl. Opt. 2011, 50, 3658–3665. [CrossRef] [PubMed]
62. Marino, F.; Distante, A.; Mazzeo, P.L.; Stella, E. A real-time visual inspection system for railway maintenance:
Automatic hexagonal-headed bolts detection. IEEE Trans. Syst. Man Cybern. 2007, 37, 418–428. [CrossRef]
63. Gibert, X.; Patel, V.M.; Chellappa, R. Robust fastener detection for autonomous visual railway track inspection.
In Proceedings of the 2015 IEEE Winter Conference on Applications of Computer Vision (WACV), Waikoloa,
HI, USA, 5–9 January 2015; pp. 694–701.
64. Singh, M.; Singh, S.; Jaiswal, J.; Hempshall, J. Autonomous rail track inspection using vision based system.
In Proceedings of the 2006 IEEE International Conference on Computational Intelligence for Homeland
Security and Personal Safety, Alexandria, VA, USA, 16–17 October 2006; pp. 56–59.
65. Li, Y.; Trinh, H.; Haas, N.; Otto, C.; Pankanti, S. Rail component detection, optimization, and assessment for
automatic rail track inspection. IEEE Trans. Intell. Transp. Syst. 2014, 15, 760–770.
66. AutoScan Report Summary [Internet]. European Union. 2019. Available online: https://cordis.europa.eu/
project/rcn/203338/reporting/en (accessed on 25 May 2019).
67. RailPod, Inc. RailPod In Action [Internet]. 2015. Available online: https://rail-pod.com/ (accessed on
22 July 2019).
68. Liu, D.K.; Dissayanake, G.; Manamperi, P.B.; Brooks, P.A.; Fang, G.; Paul, G.; Webb, S.; Kirchner, N.;
Chotiprayanakul, P.; Kwok, N.M.; et al. A robotic system for steel bridge maintenance: Research challenges
and system design. In Proceedings of the 2008 Australasian Conference on Robotics and Automation;
Australian Robotics and Automation Association, Canberra, Australia, 3–5 December 2008.
69. Wang, R.; Kawamura, Y. A magnetic climbing robot for steel bridge inspection. In Proceedings of the 11th
World Congress on Intelligent Control and Automation (WCICA), Shenyang, China, 29 June–4 July 2014;
pp. 3303–3308.
70. Oh, J.-K.; Lee, A.-Y.; Oh, S.M.; Choi, Y.; Yi, B.-J.; Yang, H.W. Design and control of bridge inspection robot
system. In Proceedings of the 2007 International Conference on Mechatronics and Automation, Harbin,
China, 5–8 August 2007; pp. 3634–3639.
71. Aoshima, S.; Chikamori, T.; Shiraishi, M. Automatic deterioration monitoring system for tunnel wall using
fusion sensors on a train-relation between train velocity and cross-correlation coefficient of interpolated
crack images. In Proceedings of the 2002 IEEE International Conference on Industrial Technology, Bankok,
Thailand, 11–14 December 2002; pp. 416–420.
Computers 2019, 8, 56 16 of 16

72. Balaguer, C.; Montero, R.; Victores, J.; Martínez, S.; Jardón, A. Towards fully automated tunnel inspection:
A survey and future trends. ISARC. In Proceedings of the International Symposium on Automation and
Robotics in Construction, Sydney, Australia, 9–11 July 2014; IAARC Publications; p. 1.
73. Montero, R.; Victores, J.; Martinez, S.; Jardón, A.; Balaguer, C. Past, present and future of robotic tunnel
inspection. Autom. Constr. 2015, 59, 99–112. [CrossRef]
74. Robotics IND, Inc. Revolutionary Robotics Autonomy [Internet]. 2013. Available online: http://www.
roboticsinventions.com/productsB2BAuto.jsp (accessed on 22 July 2019).
75. Flammini, F.; Pragliola, C.; Smarra, G. Railway infrastructure monitoring by drones. In Proceedings of the
2016 International Conference on Electrical Systems for Aircraft, Railway, Ship Propulsion and Road Vehicles
& International Transportation Electrification Conference (ESARS-ITEC), Toulouse, France, 2–4 November
2016; pp. 1–6.
76. Ranganathan, P.; Olson, E. Automated safety inspection of grade crossings. In Proceedings of the 2010
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Taipei, Taiwan, 18–22 October
2010; pp. 2149–2154.
77. Sato, F.; Nishii, T.; Takahashi, J.; Yoshida, Y.; Mitsuhashi, M.; Nenchev, D. Experimental evaluation of a
trajectory/force tracking controller for a humanoid robot cleaning a vertical surface. In Proceedings of the
2011 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), San Francisco, CA, USA,
25–30 September 2011; pp. 3179–3184.
78. Mihai, D.; Chihaia, I.A.; Surugiu, M.C.; Minea, M. Preventive Maintenance of the Railway Infrastructure
employing Robotized Platform and Virtual Instrumentation. In Proceedings of the 2018 10th International
Conference on Electronics, Computers and Artificial Intelligence (ECAI), Iasi, Romania, 28–30 June 2018;
pp. 1–6.
79. Kolberg, D.; Zühlke, D. Lean automation enabled by industry 4.0 technologies. IFAC Pap. 2015, 48, 1870–1875.
[CrossRef]
80. Durazo-Cardenas, I.; Starr, A.; Turner, C.J.; Tiwari, A.; Kirkwood, L.; Bevilacqua, M.; Tsourdos, A.; Shehab, E.;
Baguley, P.; Xu, Y.; et al. An autonomous system for maintenance scheduling data-rich complex infrastructure:
Fusing the railways’ condition, planning and cost. Transp. Res. Part C Emerg. Technol. 2018, 89, 234–253.
[CrossRef]
81. Guler, H. Optimisation of railway track maintenance and renewal works by genetic algorithms. Graevinar
2016, 68, 979–993.
82. Nikolaidis, S.; Hsu, D.; Srinivasa, S. Human-robot mutual adaptation in collaborative tasks: Models and
experiments. Int. J. Robot. Res. 2017, 36, 618–634. [CrossRef]
83. Zhang, C.; Zang, X.; Leng, Z.; Yu, H.; Zhao, J.; Zhu, Y. Human–machine force interaction design and control
for the HIT load-carrying exoskeleton. Adv. Mech. Eng. 2016, 8, 1687814016645068. [CrossRef]
84. Ackerman, E. IHMC’s ATLAS Robot Learning to Do Some Chores; IEEE Spectrum Blog: Piscataway, NJ,
USA, 2016.
85. Chestnutt, J.; Lau, M.; Cheung, G.; Kuffner, J.; Hodgins, J.; Kanade, T. Footstep planning for the honda
asimo humanoid. In Proceedings of the 2005 IEEE International Conference on Robotics and Automation,
Barcelona, Spain, 18–22 April 2005; pp. 629–634.

© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
Reproduced with permission of copyright owner. Further reproduction
prohibited without permission.