You are on page 1of 14

Railway and ICT (Information and Communication Technology)

Innovation in Railway Maintenance utilizing


Information and Communication Technology
(Smart Maintenance Initiative) Mitsunobu Takikawa

Introduction With major changes to global society using ICT,


Japan’s railways too must change proactively to become
Information and communication technology (ICT) is the transportation mode chosen by society in this new
advancing at a tremendous pace in recent years, and is environment. The Research and Development Center of the
set to change the social environment and composition JR East Group has proposed a ‘Smart Maintenance Vision’
of industry. The ‘Industry 4.0’ national project started so as not to fall behind in these changes, and is working
in Germany in April 2013 is attempting to create a fourth on R&D for innovation of railway maintenance. This article
industrial revolution to change the composition of Germany’s introduces trends in ICT and the Smart Maintenance Vision
manufacturing industry using the Internet of Things (IoT) and along with case examples of R&D currently under way.
‘big-data’ technology. In March 2014, the Industrial Internet
Consortium was launched by the private sector in the USA Recent Social Conditions in Japan
with participation of big companies, such as GE, IBM, Intel,
Cisco, and AT&T. This plan also uses IoT and big data- Japan’s population is already declining. Its industry
related technologies to make industry as a whole smarter, experienced many past successes in high - qualit y
and maintenance more efficient. monozukuri (manufacturing) by mass-production amidst the
post-war population increase
and high-economic-growth era.
But changes to the economic
structure with falling population,
Figure 1  Change of Business Model diversif ying values, and
globalization, mean conventional
ways of doing business no
longer apply, and not even large
corporations can rest easy.
Companies that are growing
re c e n t l y a re c re a t i n g n e w
business models using the latest
ICT. For example, a manufacturer
Manufacture and sell products of photocopiers successfully
switched from manufacturing
and selling photocopiers to a
new business leasing model,
making large profits (Figure 1).
This new business model was
successful because advances
in ICT allow the company to
identify the operational status
of individual photocopiers at
Collect fees according to usage and maintenance low cost, enabling maintenance
before failures.
This can be said to be the

Japan Railway & Transport Review No. 67 • Mar 2016 22


Railway and ICT (Information and Communication Technology)

Figure 2  Low-maintenance Track

Nonwoven fabric

Cement-filled layer Wide sleeper

result of changing from tangible manufacturing to intangible System) receivers and accelerometers, in addition to
services, utilizing ICT at the core. telephone and email functions. It is no exaggeration
Looking at railway maintenance from the same to say they are full-fledged all-purpose sensors with
perspective, the function of conventional R&D has been to communications functions, and it is amazing that such
break from 3D (dangerous, difficult, dirty) work and reduce devices can be purchased for less than ¥10 0,0 0 0
equipment maintenance costs. For example, in the area of (approximately US$850).
tracks, low-maintenance tracks (Figure 2) and the like have IoT is already being used for purposes such as
been developed to mechanize inspection and repair so as to monitoring aircraft flights and automobile status. For
reduce the need for maintenance. This has reduced failures example, in the aviation industry, sensors throughout the
and human errors using a tangible approach. However, aircraft monitor the status during flight. The large volumes
railways face a major change in the social environment, so of data are analyzed and used to identify signs of failures in
maintenance must change too. Looking at case examples in advance. In railways, embedding sensors in rolling stock and
other industries, intangible approaches will be needed. mechanical equipment would enable similar maintenance.

Latest Trends in ICT Big Data Analytics


Big-data analytics has been used in recent years in areas
This section covers trends in ICT and discusses ideas for such as online shopping recommendation functions and
future railway maintenance. social network analysis. In IT companies and among
experts, ‘big data’ means more than just lots of data, and
Internet of Things (IoT) is interpreted as data leading to knowledge that is helpful
IoT is a method for enabling ‘things’ to exchange information in corporate management and business and in people’s
via the Internet. In the future, we may exchange information lives. Big data includes diverse types as shown in Figure
based on data gathered via sensors while being unaware 3, including text information typified by ‘tweets’ and similar
that the other side is a machine; Machine to Machine (M2M) social networking services (SNS), photos and videos,
is another term with similar meaning but the IoT concept logs generated when accessing websites and the like,
includes people as ‘things’. The main purpose of M2M is and information from various sensors. The data volume
to increase efficiency by automatic control of machines, amounted to 2.8 Zettabytes (ZB) (2.8 × 1021 bytes) in 2012,
aiming for a ‘smart’ society. IoT has come to the focus of and is expected to grow 14-fold to 40 ZB in 2020.
attention due to the increase in devices and sensors that In addition to services, there are expectations for big
can be carried by people and the creation of a ubiquitous data technologies in the maintenance field. The problem
communications environment. For example, smartphones of aging infrastructure built during Japan’s high economic
have many sensors, such as GPS (Global Positioning growth era has come into focus, and the Ministry of

23 Japan Railway & Transport Review No. 67 • Mar 2016


Figure 3  Future Data Volumes

Smartphones 2020
40 Zettabytes
E-money Tablets
5247
Tweets Gigabytes
per person
2012
Videos 2.8 Zettabytes
90% to 95% 5% to 10%
Photos Locations
Temporary Permanent
data
Sensors stored data
Emails (discarded)

POS Systems

Internal Affairs and Communications (MIC) estimates that on a network without being taught by humans. In other
the economic impact of preventive maintenance on road words, AI became able to learn on its own without relying on
bridges will be ¥270 billion. Bridges, tunnels, and other humans to build knowledge. This Deep Learning technology
structures are maintained for railways as well because they is used not only in image recognition, but also as voice
are part of the social infrastructure. Like roads, applying recognition technology in smartphones and tablets.
big-data analytics technologies to maintenance work should The knowledge of AI including Watson and Ponanza is a
bring about new industries and help make maintenance database where past knowledge is accumulated. In railway
more efficient. maintenance, past data gained from periodic inspections
and reports of incidents is stored on many systems. If this
Artificial Intelligence (AI) data could be utilized effectively, there is a high possibility
Research on AI started in the 1950s, and rule-based expert that we could build AI with abilities greater than veteran
systems were in the spotlight in the 1980s. These expert engineers.
systems were an attempt to express the complexity of the
real world by rules alone, but they ended in failure. Reflecting Environment Surrounding Railway Business
the lessons learned, current AI has become closer to the
thought patterns of humans by ignoring rules for data input T h i s s e c t i o n d e s c r i b e s i s s u e s re l a t e d t o f u t u re
and speculating on results by statistical processing. Recent maintenance in terms of the environment surrounding JR
examples of AI are Watson, which beat the champion of a East’s railway business. First, we organize issues from the
famous American TV quiz programme in 2011, and Ponanza, standpoint of maintenance costs, and second, from the
which beat a professional shogi (Japanese chess) player in makeup of personnel.
an official match in 2013.
Current AI systems are based on machine learning Railway Maintenance Expenses
where humans teach answers. That machine learning Figure 4 shows the changes over time in railway equipment
evolved further with Deep Learning developed by Stanford maintenance costs and Figure 5 shows the changes in
University and Google in the USA in 2012. Deep Learning capital investment. Maintenance costs have not changed
successfully obtained the concept of a ‘cat’ from images much since 1988 after the privatization of Japanese National

Japan Railway & Transport Review No. 67 • Mar 2016 24


Railway and ICT (Information and Communication Technology)

Figure 4  Maintenance Costs

(billon yen)
2000

1800

1600

1400

1200
Cost

Operating cost
1000

800

600

400
Maintenance cost
200

0
87

89

91

93

95

97

99

01

03

05

07

09

11

13
19

19

19

19

19

19

19

20

20

20

20

20

20

20
Financial year

Figure 5  Investment Costs

(billon yen)
450

Others
400
Lifestyle services
350 Station services

System change
300
Investment cost

250 Better transport

200

150

Safety
100 Renewal

50

0
87

93

95

97

99

01

03

05

07

09

11

13
89

91
19

19

19

19

19

20

20

20

20

20

20

20
19

19

Financial year

25 Japan Railway & Transport Review No. 67 • Mar 2016


Railways (JNR), making up between 25% and 33% of overall equipment. Figure 6 shows the age composition for
railway operating costs. Overall railway capital investment all personnel. Veteran experts age 50 or older who bear
has been around ¥200 billion since privatization, but has a central role in railway operation make up 43% of all
been increasing in recent years. Capital investment declined employees, and they will be retiring within 10 years. Almost
in 2011 as a result of the Great East Japan Earthquake the same trend can be seen in maintenance, with few people
and tsunami. The amount for replacement of dilapidated in the mid-level segment in their 40s, making passing-on
infrastructure-related equipment is included in the item skills to new hires a pressing issue.
for safety and renewal. However, new maintenance-free
equipment in the future is being introduced at equipment Smart Maintenance Initiative
updates, so those expenses are included in the system
change item. The low-maintenance track shown in Figure 2 The Research and Development Center of JR East
is an example. Group has proposed the Smart Maintenance Initiative to
However, in current society with a decreasing population, create innovation in future railway maintenance from the
there is no guarantee that today’s expense structure and perspectives of major changes in Japan’s social environment
capital investment amounts can be maintained. With the with falling population, rapid progress in ICT, and the
anticipated fall in the number of passengers (and resultant management environment. The Smart Maintenance Initiative
decrease in railway revenues) coupled with increases in is not a tangible approach to mechanizing maintenance
personnel costs as a result of a workforce shortage, the time work and enhancing equipment. Instead, it is an intangible
to develop a new maintenance structure is coming. approach of proposing a new structure for maintenance.
The Smart Maintenance Initiative is made up of the four
Age Composition of Personnel challenges shown in Figure 7: achieving condition based
JR East had a total of 59,240 employees at 1 April 2014. maintenance (CBM), introduction of asset management,
Of these people, approximately 20% (11,530) are involved work support by AI, and database integration. The following
directly in maintenance of rolling stock and wayside explains these four challenges.

Figure 6  Age Composition of Personnel

Age: –19 .... 1%

Age: 20–24 .... 7%


Age: 55– .... 23%

Age: 25–29 .... 14%

Age: 30–34 .... 14%

Age: 50–54 .... 20%


Age: 35–39 .... 13%

Age: 45–49 .... 2%


Age: 40–44 .... 6%

Japan Railway & Transport Review No. 67 • Mar 2016 26


Railway and ICT (Information and Communication Technology)

Figure 7  Four Pillars of ‘Smart Maintenance Initiative’

Achieving Condition Based Maintenance Work Supported by Artificial Intelligence

Monitoring with commuter train


A turnout is
not switched!
Check point?

You should
check
Gathering lot of ………
inspecting data
Repair at best timing

Limit level Decision-making Knowledge Experience


based on data
Parameters

analysis and
prediction

Rules
Carrying out 'smart' Inspected Data
repairwork based on History
previous results Environment
-----
Time

Tracing after repair

Introduction of Asset Management Introduction of Integrated Database

Visualization costs and level of service


Analysis by entire maintenance system
Strategic budget planning

Condition (risk) Suggestion for new maintenance method!

Past
Future
Independent
Cost

DB

Risk Cost

Linked
DB

Legend
Clarify relation between risk and cost M = Maintenance C = Civil Engineering E = Electrical Engineering
S = Signalling R = Rolling Stock

27 Japan Railway & Transport Review No. 67 • Mar 2016


Figure 8  Time Based Maintenance and Condition Based Maintenance

Present (Time Based Maintenance) Future (Condition Based Maintenance)

Measuring Irregularity Irregularity and Track Material


By inspection car (every 3 months) Inspection by monitoring equipment installed on
commercial train (daily)
Inspection for Track Material
By visual inspection (annually)

Warning in real time


Inspection and maintenance based on Planned maintenance based on projection
standards

Jan Apr July


Jan Apr July Decision-making
Condition monitoring
Periodic inspection Standards support system

Achieving Condition Based Maintenance (CBM) at set intervals. In the example of tracks, obtaining track
Achieving Condition Based Maintenance means changing displacement data from trains in operation would allow such
the basis of maintenance from Time Based Maintenance displacement data to be obtained every day. Analyzing
(TBM) to Condition Based Maintenance (CBM). This that data would allow detailed identification of the speed at
entails a major change in the maintenance philosophy. The which track deteriorates per 1 m (per track environment).
difference between the two is explained as follows using an As a result, track deterioration could be predicted, allowing
example of maintenance for tracks, a typical type of railway repairs at the optimum timing. Decision-making based
equipment (Figure 8). Conventionally, in TBM, inspections on predictions would be a fundamental change in railway
are conducted at regular cycles (once every 3 months for maintenance.
narrow-gauge track) to obtain data on deformations (track Moreover, if displacement data could be obtained every
irregularity). Track repairs are conducted when a threshold day, the effects of repairs would be instantly evident. In other
decided based on experience (for example, 23 mm) is words, with CBM, the cycle shown in Figure 9 could be
exceeded. That threshold is decided based on maximum followed on a daily basis. That cycle entails obtaining data,
deterioration taking into consideration track irregularity (for analyzing data, making decisions, conducting repairs, and
example, 40 mm) at which derailment could occur based on evaluating the results (obtaining data). The more data is
past data and the 3-month inspection cycle. Inspections are accumulated, the smarter the important decision-making
assumed to be performed at a set cycle, so the threshold is process becomes. Conversely, in TBM, justification for
uniform regardless of the track environment and a sufficient decision-making is prescribed by rules (internal regulations,
leeway must be set. etc.), so there is inevitably little awareness about responding
Conversely, CBM is based on monitoring the status of flexibly according to budget status, etc. In fact, maintenance
facilities and equipment rather than performing inspections criteria for track repair have not changed in about 50 years.

Japan Railway & Transport Review No. 67 • Mar 2016 28


Railway and ICT (Information and Communication Technology)

Figure 9  Condition Based Maintenance Cycle

Evaluation of repair work Gathering various inspection


data by monitoring, etc.
Evaluating timing method of repairs based on deterioration
after repairs

Evaluation Data
gathering

Smart repairs based on Forecast of deterioration


past repair results Data
Repair base Data
work analysis

Deterioration
Decision
making Time

- Maintenance at optimal intervals Cost


Min Cost Min Risk

- Projected maintenance
- Minimized accident risk and Min Total
maintenance cost
Risk

Cost

Introduction of Asset Management Work Support by AI


Asset management is the concept of considering The purpose of work support by AI is to back-up evaluation
equipment as assets and efficiently managing them from work currently done by maintenance engineers. Here, AI
a long-term view. There are many case studies on this for is more than a support system using unstructured data,
bridges, tunnels, and other civil-engineering structures such as text and images; it can be applied to discovering
where deterioration is slow and the scale of repair work new relationships beyond human experience by big-data
is large. However, there are currently few examples as an analysis using structured data.
actual business practice. A possible reason is because The following introduces a support method for when
identifying equipment and evaluating deterioration over the equipment failure occurs as a past case example using
long term are difficult. If these issues could be overcome text data. When equipment failure occurs, maintenance
by smart maintenance, maintenance engineers could engineers need to quickly identify the cause of the failure.
compare various repair methods and periods based on Experienced engineers can accurately infer the failure cause
the deterioration status of equipment. This would enable based on past experience, the status of equipment, and
optimum repair plans to be proposed over a long-term various environmental conditions to quickly recover from
span. Like CBM, this concept is thought to be a method of the failure. However, it is difficult for engineers with little
supporting decision-making by maintenance engineers. experience to identify the cause of a failure. If we create a

29 Japan Railway & Transport Review No. 67 • Mar 2016


Figure 10  Integrated Platform for Smart Maintenance

Integrated Platform for Smart Maintenance

Database for analysis


Dashboard for executive
Support system for
maintenance This platform supports various kind of work!

Enterprise service bus Data can be integrated while


existing systems work.

Independent Systems Social Data

Infrastructure data Signal data

Rolling stock data Track data

SNS
Electric power data Security system data

Operation data Failure data

database of the tacit knowledge and experiences of veteran data is handled freely. JR East currently has various systems
engineers and utilize AI-related technologies, we should be for individual divisions. These were established to optimize
able to infer the cause of failures. divisional work, so each has a completely independent
system structure; the individual systems are not linked, and
Database Integration there is no common data strategy.
To make CBM a reality, introduce asset management, and To make quick management decisions and provide
achieve work support by AI, we need an environment where better support to front-line divisions, it is important to

Japan Railway & Transport Review No. 67 • Mar 2016 30


Railway and ICT (Information and Communication Technology)

Figure 11  Monitoring Train: Series E235

Catenary monitoring system

Control center Maintenance


depot

Operation control

Trouble prediction

Engineer onsite

Trouble recovery

Track monitoring system

manage the vast amount of data held by the company in Monitoring Train: Series E235
a centralized manner. In other words, we need to integrate Figure 11 shows Series E235 commuter rolling stock
the separate internal systems and construct a large system manufactured in 2015 for Tokyo’s Yamanote Line. This is
(common platform based on individual databases) that can the world’s first rolling stock to monitor its own onboard
also link with external data (weather information and SNS devices as well as wayside equipment (track equipment and
data). Figure 10 shows this concept. Using this platform power equipment) while operating. Monitoring data captured
everyone from front-line organizations to management will onboard is sent to the wayside by wireless communications
be able to use the same data, allowing quick and accurate and distributed to dispatch and maintenance worksites. Data
decision-making in the company. received by an individual worksite is analyzed and can be
used to plan later maintenance.
Efforts in R&D One example of monitoring data analysis is track
irregularities captured by a track monitoring system on other
We are conducting various R & D to achieve smar t rolling stock. Data for the same location is captured up to
maintenance. Three case studies are introduced here. five times. Figure 12 shows the change in track irregularities
after repairs. We can see that track subsidence after

31 Japan Railway & Transport Review No. 67 • Mar 2016


Figure 12  Analyzed Monitoring Data (Track Irregularity)

Machine tamping
(mm)
0
-2
-4
-6
-8
-10
-12 Repair
-14
-16
-18
-20
2013/06/26 2013/09/25 2013/12/25 2014/03/26

Manual tamping
(mm)
0
-2
-4
-6
-8
-10
-12
-14 Repair
-16
-18
-20
2013/06/26 2013/09/25 2013/12/25 2014/03/26

Figure 13  Example of Maintenance Support System

Catenary height

Track irregularity (right)

Track irregularity (left)

Signal facilities

Japan Railway & Transport Review No. 67 • Mar 2016 32


Railway and ICT (Information and Communication Technology)

Figure 14  Point Cloud Data with Mobile Mapping System department. We first built a prototype
virtual environment and support tool
w here equipm ent and inspe ctio n
info r m atio n separated by syste m
are displayed on a single screen to
extract issues to solve. Figure 13 is a
case study of a chart where power
and track maintenance equipment is
inspected and the location status of
communications equipment is displayed
as a list. This study showed that location
information and time information are
impor tant in co ordinating railway
maintenance. Some equipment does
not have location information, and we
confirmed the necessity of providing it
efficiently.
We are considering use of
Figure 15  Heat Map Showing Track Condition History G e o gra p hic Info r m ati o n S yste m s
(GIS) already introduced by railways
Present in other countries as a method for
35 18 3 11 31 25 25 29 25 27 14 23 34 31 41 35 38 22 29 31 42 36 24 17 efficiently managing equipment location
32 4 2 10 29 24 23 22 21 14 7 13 25 17 31 21 31 14 40 48 53 56 44 38 information. Coordinates (longitude and
33 34 2 10 27 19 22 34 27 33 17 35 43 39 43 40 36 26 41 44 44 55 43 34
Bad latitude) are necessary to use GIS. For
32 32 2 7 25 15 20 32 28 29 17 32 39 37 45 35 34 25 38 44 44 55 37 27
31 32 1 7 25 16 17 33 28 28 18 30 40 37 45 41 35 26 36 44 44 57 50 36 this reason, we need to build a database
28 27 1 3 23 13 23 32 27 26 20 28 37 36 43 43 35 26 40 39 44 53 47 33 of coordinate information for equipment
20 25 2 0 21 12 25 38 38 28 26 30 45 39 54 54 40 36 42 31 16 23 22 9 without location information. A possible
21 22 1 1 21 9 24 34 34 23 23 26 43 38 51 53 35 25 42 29 49 42 33 37
solution is to use Mobile Mapping
17 00 Good
0 1 17 8 21 40 25 29 24 25 35 31 39 58 36 17 27 13 42 26 40 26
20 21 1 1 14 5 21 39 23 28 23 20 31 34 37 60 34 16 32 11 42 26 39 23 S y s te m s ( M M S ) . W i t h M M S , t h e
15 20 2 0 14 5 20 40 26 26 23 17 29 25 35 48 36 13 27 9 37 18 37 23 surrounding trackside equipment can
14 20 2 0 13 Bad
4 19 39 25 25 19 14 28 25 35 46 36 12 29 7 37 23 34 21
be acquired as 3D data while running a
8 10 0 0 9 3 16 39 23 24 17 10 26 24 31 44 31 11 25 5 32 18 35 20
19 22 6 0 18 7 17 22 21 18 23 17 23 21 29 46 33 10 26 2 32 15 30 16 maintenance vehicle. Figure 14 shows
20 22 6 0 15 7 22 29 24 35 28 25 33 36 50 48 41 25 29 19 45 30 43 32 an example of an image obtained by
19 17 5 0 15 6 22 26 23 35 28 24 34 32 45 42 40 21 27 12 41 29 41 31
measurements. Although the figure
48 months Location (100-m intervals) looks like a photograph, it is actually a
ago
collection of points, and each point has
coordinate information. Using this data,
we can efficiently obtain coordinate
information about equipment.

manual repairs occurs earlier than after mechanical repairs. Passing-on Skills
By constantly conducting such analysis, we can discover The know-how of veteran engineers is very important when
deterioration details by location according to differences conducting maintenance. However, these veterans in their
in repair method and environment, supporting efficient 50s or older, who make up about 40% of all employees,
maintenance. are approaching retirement and their know-how must be
passed-on to younger engineers. We have started studying
Platform for Big-Data Analytics the possibility of using amassed maintenance data for
As much data as possible must be collected for big-data smoothly passing-on skills.
analytics and analysis by AI. We are studying a method The first case study is a method of expressing past
to integrate inspection and specification data for wayside information so that it can be used in decision-making work
equipment managed by independent systems in each (Figure 15). This expresses the status of track irregularities

33 Japan Railway & Transport Review No. 67 • Mar 2016


Figure 16  Image Recognition using Machine Learning

Input image

Surface
Machine
state
learning
classifier

Head checks Corrugation Squat Other defects

Japan Railway & Transport Review No. 67 • Mar 2016 34


Railway and ICT (Information and Communication Technology)

every 3 months for 4 years as a heat map. The horizontal


axis is the location on the line (per 100 m) and the vertical
axis shows the track status history. The darker locations
show where the deterioration status is good (the actual
colour is green) or poor (the actual colour is red). By viewing
this figure, the engineer can see the track deterioration
status and changes at a glance, so it can be used as a
reference in repair planning.
The second case study is image analysis using machine
learning, a form of AI technology (Figure 16). This is an
attempt to use images showing the rail head obtained by
track monitoring devices to discover faults based on the
experience of veteran engineers. This image processing
implements machine learning and can reflect the know-how
of veteran engineers. Even using only part of the data, rail
head defects could be discovered with a success rate of
about 98%.

Conclusion
The essence of maintenance —accurately perceiving
deterioration and deciding the optimal timing and method
for repairs— can be strongly supported by the Smart
Maintenance Initiative, making CBM a reality, introducing
asset management, and supporting work using AI.
Furthermore, integration of databases of fers a new
mechanism for making the most of data that is now ‘buried’
in the system and is not being used effectively.
I strongly believe that achieving smart maintenance will
lead to major reforms in railway maintenance.

References
Atsushi Yokoyama, ‘Innovation in Railways: Platform for Maintenance
Innovation Using ICT’ JR East Technical Review No. 29 (summer 2014)

Mitsunobu Takikawa, ‘Impact of ICT Trends on Future Railway Maintenance’


JR East Technical Review No. 29 (summer 2014)

Mitsunobu Takikawa
Dr Takikawa is a chief researcher at Technical Center, Research and
Development Center of JR East. He joined JR East in 1990. He earned
his doctorate in mechanical engineering from Tokyo University of
Science in 2013.

35 Japan Railway & Transport Review No. 67 • Mar 2016