An Intelligent Framework and Prototype for Autonomous

Maintenance Planning in the Rail Industry

C. J. Turner

, A. Tiwari

, A. Starr

, I. Durazo-Cardenas

and K. Blacktop

Manufacturing and Materials Department, Cranfield University, Bedford, U.K.

Network Rail, Milton Keynes, U.K.

Keywords: Rail Industry, Computational Intelligence, Data Mining, Data Fusion, Autonomous Operation, Planning.

Abstract: This paper details the development of the AUTONOM project, a project that aims to provide an enterprise

system tailored to the planning needs of the rail industry. AUTONOM extends research in novel sensing,

scheduling, and decision-making strategies customised for the automated planning of maintenance activities

within the rail industry. This paper sets out a framework and software prototype and details the current

progress of the project. In the continuation of the AUTONOM project it is anticipated that the combination

of techniques brought together in this work will be capable of addressing a wider range of problem types,

offered by Network rail and organisations in different industries.

1 INTRODUCTION

Recent developments in both Computational

Intelligence and web based information systems

have made possible a new era of interconnected

decision support platforms capable of autonomous

operation. The AUTONOM project provides such an

enterprise system tailored to the planning needs of

the rail industry. AUTONOM aims to extend

research in novel sensing, scheduling, and decision-

making strategies customised for the automated

planning of maintenance activities within the rail

industry. The case study for this work is UK rail

infrastructure provider Network Rail (Network Rail

is the organisation that is responsible for maintaining

and developing the UK rail infrastructure including

signalling, bridges, tunnels, level crossings, viaducts

and 17 key stations within the country).

Maintenance of widely-dispersed assets is expensive

because it involves widespread inspection, checking

and measurement. The integration of sensor-based

information in geographically dispersed and less

structured environments poses challenges in

technology and cost justification. Academic

challenges include improvement of embedded

sensing, reliable estimation of monitoring

parameters, a unified approach to the mathematics

and data structures, and a rigorous approach to cost

estimation and benefit analysis. The industrial

drivers include standardisation, automation,

connectivity, and reduction of unit cost (Starr et al.

2013). The overall aim is to enable improved

integration between architectural levels in data-rich

environments in automated, intelligent maintenance,

responding to business pull and demonstrating value.

A prototype is currently being developed by the

authors that will deliver this vision as a web enabled

platform capable of delivering automated decision

capability to managers involved in rail maintenance

planning activities.

2 RELEVANT RESEARCH

On embarking on such a project it was necessary to

investigate current approaches to planning and cases

of autonomous operation in the rail industry. In

addition to literature a think tank group was

convened to examine areas of railway operation that

would benefit from automation. The think tank

group was made up of senior representatives of

Network Rail and other corporate partners of the

wider AUTONOM project. The four areas shown in

Figure 1 were highlighted by this group as having

potential for automation and are considered as

priority areas by Network Rail and the authors.

Turner C., Tiwari A., Starr A., Durazo-Cardenas I. and Blacktop K..

An Intelligent Framework and Prototype for Autonomous Maintenance Planning in the Rail Industry.

DOI: 10.5220/0005462800830089

In Proceedings of the 1st International Conference on Vehicle Technology and Intelligent Transport Systems (VEHITS-2015), pages 83-89

ISBN: 978-989-758-109-0

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

Figure 1: Planning and scheduling activities relevant to the rail industry.

2.1 Maintenance in the Rail Industry

As mentioned by Dadashi et al. (2011) there has

been a move from reactive to diagnostic and

prognostic modes of railway maintenance; where

fault diagnosis and prediction are key. As part of this

movement work has been conducted into the area of

fault states and the interdependencies between the

states in Schöbel and Maly (2012). This paper makes

clear the need for active monitoring of rail vehicles

so logical connections between fault states can be

made. In the work of Dadashi et al. (2012) data

from assets deployed in the field can increasingly be

made available to mobile devices. Bye (2013) details

the use of mobile devices and the design of

application interfaces with case studies drawn from

Network Rail. The subject of risk centred

maintenance is investigated by Selvik and Aven

(2011). These authors make the point that Reliability

Centred Maintenance (RCM) for preventative

maintenance practice is well understood in industry

though it does present limitations when used to

quantify risk and uncertainty in projects. The

importance of organisational design in railway

maintenance operations is highlighted by Jiang et al.

(2012) Maintenance operations may be performed

by a collection of co-operating organisations, having

a model of how such organisations operate together

is in the opinion of Jiang et al. (2012) essential.

2.2 Planning in the Rail Industry

Timetable planning is an active area of research for

the rail industry. The work of Yang et al. (2010)

investigates the development of incomplete cyclic

timetables for trains. Real time re-scheduling of

trains is the subject of Wegele et al. (2010), where

genetic algorithms are used to perform a heuristic re-

ordering of trains when delays are encountered in

the rail network. The mark‐up language RailML is

also mentioned in this work. RailML is an XML

based metadata language that defines a set of

common rail specific terminology as meta-

descriptions (RailML is detailed in Nash et al.

(2004)). The timetabling problem is framed by Ho et

al. (2012), as one of multi objective‐optimisation in

that a feasible solution must be identified within a

limited time constraint that satisfies a number of

objectives. Scheduling of trains in terms of energy

efficiency goals is another active area of interest and

one such approach is described in the work of Hu et

al. (2013). This approach utilises a combination of a

standard multi‐objective optimisation algorithm and

a fuzzy multi‐objective optimisation approach to

find train allocations and movements that balance

fuel efficiency with passenger travel time. In Peng et

al. (2013) the optimal routing and scheduling of

periodic inspections in a rail network are explored in

VEHITS2015-InternationalConferenceonVehicleTechnologyandIntelligentTransportSystems

the context of long term planning. A heuristic

algorithm is put forward as part of this research that

utilises a local search method allowing for the

scheduling problem to be divided into sub sets, as in

similar work by Peng et al. (2012). Improved results

are found over existing manual techniques although

this approach is only able to add new tasks at the end

of the schedule rather than within the schedule.

2.3 Sensor Fusion

One of the major developments in train control

systems in recent years has been the European Rail

Traffic Management System (ERTMS). The

ERTMS system is composed of two main entities,

the European Train Control System (ETCS) and

Global System for Mobile Communications-Railway

(GSM-R) (Abed, 2010). ETCS is comprised of two

main components, the Automatic Train Protection

System (ATP) (automated emergency braking at red

signals) and signalling system (in cab signalling).

GSM-R is an international standard for mobile voice

and data communication transmission in a railway

environment and primarily used for train to control

centre communication. The different levels of

ERTMS are detailed in Abed (2010). Beyond the

area of train control systems, such as ETCS, much

research has been carried out in the area of

signalling and its analysis for safety related

purposes. The work of Filip et al. (2010) examines

the use of Global Navigation Satellite Systems

(GNNS) in relation to railway safety applications. In

this study the authors examine the applicability of

existing GNSS technologies and their modes for use

in the railway industry, highlighting their current

limitations. As Filip et al. (2008) mentions satellite

technology will have a significant impact on the

railways in that the trackside sensor equipment (such

as Balaises and hot box sensors) will be superseded

by asset located sensors that determine the actual

real time position of moving assets (Rail vehicles)

via satellite. In terms of railway signalling, research

has been conducted involving areas such as satellite

technology (Filip et al., 2008, 2010) though there

has also been an investigation into the use of Internet

Protocol for the control of signals reducing the need

for the installation of traditional copper wiring

(Endo, et al., 2008).

2.4 Autonomous Systems

Autonomous systems may range from

semi‐automated to totally autonomous (without

human intervention) operation. Semi‐automated

operation includes ETCS in‐cab signalling and

automated train braking systems. Fully automated

operation examples include the automated operation

of trains. Autonomous systems can contain complex

control logic often incorporating computational

intelligence techniques such as fuzzy logic, genetic

algorithms and autonomous agents. Agents are said

to embody two principles: an ability to act

autonomously and the capability to interact with

other agents (Wooldridge, 2008). Multi-agent

systems are then, by definition, interacting

collections of agents (Wooldridge, 2008). The

subject of autonomous real‐time planning has been

investigated by Cresswell et al. (2013) in their work

on the LOCM (Learning Object‐Centred Models)

system. Automated planning has also been

investigated by Fernandez et al. (2013) who propose

an architecture for the automation of data mining

tasks. Again the PDDL (Planning Domain

Definition) language features heavily in this

approach. These papers, while not currently applied

to rail uses, do hold some potential for further

investigation with regards to autonomous rail

research. There are only limited investigations into

fully autonomous systems usually concentrating on

the automatic operation of metro/light rail systems

as in Dominguez et al. 2010) and Wackrow and

Slamen (2013). Of interest in the enablement of

autonomous operation is the use of ontology. An

example of the use of ontologies in the management

of rail systems is provided by Briola et al. (2013)

who present a semantic architecture for the

centralised control of railway traffic. One of the

main advantages of this system is the ability for a

user to make natural language queries on the data

being collected. Two key objectives of this approach

are to reduce the overall cost of managing the rail

system through autonomous or semi‐autonomous

operation and the ability to make real time changes

to train movements and their scheduling when

encountering problems that may lead to service

disruption such as faulty trains and damage within

the infrastructure (Briola et al., 2013).

Opportunities exist to review maintenance planning

and scheduling, among other activities, with a view

to autonomous or semi‐autonomous operation.

AnIntelligentFrameworkandPrototypeforAutonomousMaintenancePlanningintheRailIndustry

Figure 2: AUTONOM Framework Illustration.

3 FRAMEWORK AND

PROTOTYPE

In the development of the framework and prototype

it was decided that the area of planning and

scheduling for rail maintenance should be addressed

by the AUTONOM project. As a first stage the

framework shown in Figure 2 was developed,

outlining the major units of research that are

required to address maintenance planning activities.

3.1 AUTONOM Framework

In Figure 2 it can be seen that there are three key

areas of research that comprise AUTONOM, sensor

fusion, planning and scheduling and costing. Sensor

fusion involves the processing and analysis of raw

data provided by sensors located on assets; in the

case study (detailed later in the paper) this data will

be supplied by Network Rail. The data received will

describe the degradation state of the tracks and faults

found along the line. From this data it will be

possible to look for trends and identify triggers for

when track repairs become necessary and the

urgency of the repair. On identifying when a repair

has become urgent sensor fusion will raise an alert.

The planning and scheduling area will provide a

Gantt chart of all scheduled maintenance activities

for certain aspects of track maintenance. It will be

possible to schedule or reschedule a maintenance job

depending on the urgency of the work and the

resources available. An important component of

scheduling is of course cost. The area of costing will

also be examined by AUTONOM. It will be possible

to cost a repair in the context of it being a single

(one-off) job or in combination with other

maintenance activities. It will also be possible to

build in costing for uncertain elements such as the

effect of the weather, ground conditions and

resources available at the site of the repair and the

time window when the repair will take place. When

undertaking repairs on the rail system the time

window available is often limited by the amount of

possession time available to Network Rail (that is

the amount of time when access to the track by rail

VEHITS2015-InternationalConferenceonVehicleTechnologyandIntelligentTransportSystems

vehicles can be denied for safe completion of

maintenance activities). In some cases such as when

repairs can be made by track grinding machines it

may be possible to complete a maintenance activity

outside of track possession time. Within the

framework outlined it will be possible to pass data

between areas such as the feedback loop between

costing and planning and scheduling.

3.2 AUTONOM Prototype

Figure 3 shows a schematic of the prototype that is

currently being developed as part of AUTONOM. In

this prototype it can be seen that the three areas

illustrated in the framework (shown in Figure 2)

have now become components In addition a fourth

component has been added, the integration

component. This component will coordinate the data

being passed between the other three components

and act as a portal for the user interfaces presented

by each component. This application will make full

use of networked data (shown as 1 in Figure 3) as

many live (or near live) streams are available. Where

live data is not available as a stream data files will

be made available on Network Rail servers (updated

on a regular hourly/daily basis). As mentioned in the

framework description the sensor fusion component

(2 in Figure 3) will utilise a combination of data

mining techniques and statistical approaches to

identify when urgent track maintenance is necessary

and raise an alert to the integration component (3 in

Figure 3). The integration component sends the alert

to the planning and scheduling component (4 in

Figure 3) for the scheduling of the maintenance that

is necessary for the repair required. A further loop

through integration allows for the repair to be valued

by the costing component (6 in Figure 3). Once a

costing is available it will be possible for the

planning and scheduling component to produce an

optimised scheduling for the repair. As the data

made available by each component will accessible to

the prototype as a whole it will be possible for the

planning and scheduling component to produce a

range of scheduling scenarios each, perhaps, with

different time and cost implications.

It would then be for the user to either select the

scenario (through the interface provided by the

integration component) or set rules on how scenarios

should be chosen to allow for fully automated

scheduling to occur. The optimisation approach will

build on existing heuristics for track repairs in

combination with the use of a multiple objective

optimisation approach such as genetic algorithm

(where a population of schedules may be developed

and optimised for trade-off between cost and time).

In the development of the prototype it is envisaged

that the application should be web accessible. The

application will be built on a Microsoft platform

Figure 3: Schematic diagram of the AUTONOM Prototype.

AnIntelligentFrameworkandPrototypeforAutonomousMaintenancePlanningintheRailIndustry

utilising the C# programming language and the

integration component will provide a web based

interface. The use of web technologies also raises

the possibility of utilising the application at trackside

for scenarios when work has to be rescheduled at

short notice due to unforeseen circumstances (such

as a change in weather/ground conditions and

overruns in time taken to complete the repair). In

this case the work of Bye (2013) will be

instrumental as the interfaces provided to Network

Rail mobile devices could be utilised in the display

of the AUTONOM prototype. In terms of data

exchange it is likely that many streams (and data

files) provided by Network Rail will be in a

proprietary format. This will necessitate the data

fusion component to provide a transformation of this

data into an XML format. In this task the RailML

format (Nash et al., 2004) may be informative in

constructing a suitable XML representation.

4 CASE STUDY

As a practical demonstration of the approach

outlined in this paper the illustrated prototype

(Figure 3) is currently under development with the

UK infrastructure provider Network Rail providing a

case study for its practical use. In terms of the

maintenance activities to be analysed by

AUTONOM it is likely that there will be a

concentration on track repair activities performed by

rail grinding and tamping vehicles. The practice of

rail grinding is used into to address defects such

cracks in the rail head (the part of the rail in contact

with train wheels). In practice the top layers of metal

of the track head (containing the crack) are ground

away to the depth of the crack. Rail grinding

machines are normally deployed as rail vehicles that

can perform the grinding operation while in motion

along the length of track affected by rail head

cracks. The practice of tamping is different from

grinding in that it addresses problems in track

subsidence by lifting the track and squeezing the

ballast underneath to raise the height of the track. As

with grinding tamping machines normally take the

form of a rail vehicle that can move up and down the

affected length of track. As tamping can over time

damage the ballast underneath the track (crushing it

into a finer grain) the practice can be combined with

stone blowing where new stones can be inserted

below a section of track.

Both tamping and grinding rail vehicles contain a

range of sensors and are capable of producing data

on the state of the track and the repairs being carried

out. In addition Network Rail utilise a number of

track inspection trains that they use to monitor track

condition throughout the UK rail network. Such

inspection trains also produce track condition data in

volume.

As the AUTONOM project progresses use will be

made of these data sources in order to develop, test

and eventually validate the software prototype.

5 CONCLUSIONS

This paper has set out the current position and stage

of development of the AUTONOM project; a project

that aims to provide an enterprise system tailored to

the maintenance needs of the rail industry, with an

initial focus on that area of rail maintenance

planning. The software prototype resulting from this

project will take advantage of the recent

developments in computational intelligence and the

availability of networked data stores and feeds made

available by the case study organisation UK rail

infrastructure provider Network Rail to provide a

web based enterprise solution for maintenance

activities.

In the continuation of the AUTONOM project it is

anticipated that the combination of techniques

brought together in this work will be capable of

addressing a wider range of problem types, offered

by Network Rail and organisations in different

industries.

ACKNOWLEDGEMENTS

This research has been funded by the UK

Engineering and Physical Sciences Research

Council (EPSRC) and UK rail infrastructure

provider Network Rail.

REFERENCES

Abed, S.K., 2010. European rail traffic management

system - An overview. In: 1st International

Conference on Energy, Power and Control (EPC-IQ),

Basra, Iraq, 173-180.

Briola D, Caccia R, Bozzano M and Locoro A., 2013.

Ontologica: Exploiting ontologies and natural

language for railway management. International

Journal of Knowledge-Based and Intelligent

Engineering Systems, 17, 3-15.

Bye R., 2013. Designing Mobile User Experiences:

Disruptive Innovation in Railway Asset Information.

VEHITS2015-InternationalConferenceonVehicleTechnologyandIntelligentTransportSystems

In: Dadashi N. et al. (eds) Rail Human Factors:

Supporting Reliability, Safety and Cost Reduction,

Leiden, CRC Press, 453.

Cresswell SN, McCluskey TL and West MM., 2013.

Acquiring planning domain models using LOCM.

Knowledge Engineering Review, 28, 195-213.

Dadashi N, Wilson JR, Sharples S, Golightly D and

Clarke, T., 2011. A framework of data processing for

decision making in railway intelligent infrastructure,

2011 IEEE International Multi-Disciplinary

Conference on Cognitive Methods in Situation

Awareness and Decision Support (CogSIMA 2011),

Miami Beach, Florida, 276-283.

Dadashi Y, Sharples S, Wilson J and Clarke T., 2012.

Investigating presentation of rail-specific spatial

information on handheld computer screens. Personal

and Ubiquitous Computing, 16, 1051-1064.

Dominguez M, Fernandez A, Cucala AP and Blanquer J.,

2010. Efficient Design of Automatic Train Operation

Speed Profiles with On Board Energy Storage

Devices. In: Ning B et al. (eds) Computers in Railways

XII: Computer System Design and Operation in the

Railway and Other Transit Systems. Southampton,

WIT Press, 509 – 520.

Endo M, Okada T, Watanabe D, Aimi K, Kunifuji T and

Matsumoto M., 2008. A safety-related transmission

method for a new railway signalling system based on

an IP-Network. In: Allan J et al. (eds) Computers in

Railways XI: WIT Transactions on the Built

Environment, Southampton, WIT Press, 113.

Fernández S, De La Rosa T, Fernández F, Suárez R, Ortiz

J, Borrajo D and Manzano D., 2013. Using automated

planning for improving data mining processes.

Knowledge Engineering Review, 28, 157-173.

Filip A, Bažant L and Mocek H., 2010. The experimental

evaluation of the EGNOS safety-of-life services for

railway signalling, In: Ning B et al. (eds) Computers

in Railways XII: Computer System Design and

Operation in the Railway and Other Transit System,

Southampton, WIT Press, 735.

Filip A, Beugin J, Marais J and Mocek H., 2008. Safety

concept of railway signalling based on Galileo Safety-

of-Life Service. In: Allan J et al. (eds) Computers in

Railways XI: WIT Transactions on the Built

Environment, Southampton, WIT Press, 103-112.

Nash A, Huerlimann D, Schuette J and Krauss VP., 2004.

RailML-a standard data interface for railroad

applications. In: Allan J et al. (eds) Computers in

Railways IX, WIT Transactions on the Built

Environment, Southampton, WIT Press, 233-240.

Peng, F, Ouyang, Y., and Somani, K., 2013. Optimal

routing and scheduling of periodic inspections in

large-scale railroad networks, Journal of Rail

Transport Planning and Management, 163-171.

Peng, F, Ouyang, Y., 2012. Track maintenance production

team scheduling in railroad networks, Transport Res

B-Meth, 46, 1474–1488.

Ho TK, Tsang CW, Ip KH and Kwan KS., 2012. Train

service timetabling in railway open markets by particle

swarm optimisation. Expert Syst Appl, 39, 861-868.

Hu H, Li K, and Xu X. (2013). A multi-objective train-

scheduling optimization model considering

locomotive assignment and segment emission

constraints for energy saving, Journal of Modern

Transportation, 21, 9-16.

Jiang J, Huisman B and Dignum V., 2012. Agent-based

multi-organizational interaction design: A case study

of the Dutch railway system, Proceedings - 2012

IEEE/WIC/ACM International Conference on

Intelligent Agent Technology (IAT 2012), Macau,

China, 196-203.

Schöbel A and Maly T., 2012. Operational fault states in

railways. European Transport Research Review, 4,

107-113.

Selvik JT and Aven T., 2011. A framework for reliability

and risk centered maintenance. Reliability Engineering

and System Safety, 96, 324-331.

Starr A, Baguley P, Xu, Y Tsourdos A, Shehab, E, and

Tiwari, A. 2013 AUTONOM: Integrated through-life

support for high-value systems: Details of Grant,

http://gow.epsrc.ac.uk/NGBOViewGrant.aspx?GrantR

ef=EP/J011630/1, Visited on 17/03/15.

Wackrow J and Slamen A., 2013. The implications of

automation on human intervention at London

underground, Rail Human Factors: Supporting

Reliability. Safety and Cost Reduction, 209.

Wegele S, Corman F and D’Ariano A., 2010. Comparing

the effectiveness of two real-time train rescheduling

systems in the case of perturbed trafﬁc conditions. In:

Hanson I (ed) Timetable Planning and Information

Quality, Southampton, WIT Press, 189.

Wooldridge M., 2008. An introduction to multiagent

systems, Wiley, Chichester.

Yang D, Nie L, Tan Y, He Z and Zhang Y., 2010.

Working out an incomplete cyclic train timetable for

high-speed railways by computer. In: Ning B et al.

(eds) Computers in Railways XII, WIT Transactions

on the Built Environment, Southampton, WIT Press,

889.

AnIntelligentFrameworkandPrototypeforAutonomousMaintenancePlanningintheRailIndustry