3rd IFAC Workshop on Advanced Maintenance Engineering, Service and Technology

3rd IFAC Workshop on Advanced Maintenance Engineering, Service and Technology

October 19-21, 2016. Biarritz, France
October
3rd IFAC19-21, 2016.on
Workshop Biarritz,
AdvancedFrance Available online
Maintenance at www.sciencedirect.com
Engineering, Service and Technology
October 19-21, 2016. Biarritz, France

ScienceDirect
IFAC-PapersOnLine 49-28 (2016) 214219
Maintenance Analytics
Maintenance Analytics The
The New
New Know
Know in
in Maintenance
Maintenance
Maintenance Analytics The New Know in Maintenance
Ramin
Ramin Karim*, Jesper Westerberg**,Diego
Karim*, Jesper Westerberg**,Diego Galar*,
Galar*, Uday
Uday Kumar*
Kumar*
Ramin
** Division Karim*, Jesper Westerberg**,Diego Galar*, Uday Kumar*
Division of
of Operation,
Operation, Maintenance
Maintenance Engineering,
Engineering, Lule
Lule university
university ofof technology
technology
SE-971
SE-971 87,
87, Lule,
Lule, Sweden
Sweden ([ramin.karim]
([ramin.karim] [diego.galar]
[diego.galar] [uday.kumar]
[uday.kumar] @ltu.se)
@ltu.se)
* Division of Operation, Maintenance Engineering, Lule university of technology
SE-971 87, Lule, Sweden ([ramin.karim]
** [diego.galar] [uday.kumar] @ltu.se)
** eMaintenance365
eMaintenance365 AB AB
Aurorum
Aurorum 1C,1C, SE-977
SE-977 75,75, Lule,
Lule, Sweden
Sweden (jesper.westerberg@emaintenance365.se)
(jesper.westerberg@emaintenance365.se)
** eMaintenance365 AB
Aurorum 1C, SE-977 75, Lule, Sweden (jesper.westerberg@emaintenance365.se)
Abstract:
Abstract: Decision-making
Decision-making in in maintenance
maintenance has has to
to be
be augmented
augmented to to instantly
instantly understand
understand andand efficiently
efficiently act,
act,
i.e. the new
i.e. the new know. The new
know. The new in know
know in maintenance
in maintenance needs
needs to focus on
to focus on two aspects
two aspects of knowing:
of knowing: 1)
1) what can
what can be
be
Abstract:
known andDecision-making
2) what must be maintenance
known, in order has
to to be
enableaugmented
the to instantly
maintenance understand
decision-makers and
to efficiently
take act,
appropriate
known and 2)
i.e. the new whatThe
know. mustnew known,
be know in in order to enable
maintenance needsthe
to maintenance
focus on twodecision-makers
aspects of knowing:to take appropriate
1) what can be
actions.
actions.
known and 2) what must be known, in order to enable the maintenance decision-makers to take appropriate
Hence,
Hence, the
actions. the purpose
purpose ofof this
this paper is to
paper is to propose
propose aa concept
concept for
for knowledge
knowledge discovery
discovery inin maintenance
maintenance withwith focus
focus
on
on Big Data
Big the
Data and analytics.
and analytics. The concept
The isconcept is called
is called Maintenance
Maintenance Analytics
Analytics (MA). MA
(MA).inMA focuses
focuses inin the new
thefocus
new
Hence,
knowledge purpose of this
discovery in paper to propose
maintenance. MA a addresses
concept forthe
knowledge
process discovery
of maintenance
discovery, with
understanding, and
knowledge
on Big Data discovery
and in maintenance.
analytics. The concept MA
is addresses
called the process
Maintenance of discovery,
Analytics (MA). MA understanding,
focuses in the and
new
communication
communication of maintenance
of maintenance data from
data fromMA four time-related
four addresses
time-related perspectives,
perspectives, i.e.
i.e. 1) Maintenance
1) Maintenance Descriptive
Descriptive
knowledge discovery in maintenance. the process of discovery, understanding, and
Analytics (monitoring);
Analytics
communication(monitoring);
of
2)
2) Maintenance
maintenance Maintenance
data from
Diagnostic
Diagnostic
four
Analytics;
Analytics;
time-related
3)
3) Maintenance
Maintenance
perspectives, i.e. 1)
Predictive Analytics;
Predictive Descriptive
Maintenance Analytics;
and
and 4)4) Maintenance
Maintenance Prescriptive
Prescriptive analytics.
analytics.
Analytics (monitoring); 2) Maintenance Diagnostic Analytics; 3) Maintenance Predictive Analytics;
2016,
and IFAC (International
4) Maintenance Federation
Prescriptive of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.
analytics.
Keywords:
Keywords: big data, maintenance analytics, eMaintenance, Knowledge discovery, maintenance decision support.
big data, maintenance analytics, eMaintenance, Knowledge discovery, maintenance decision support.
Keywords: big data, maintenance analytics, eMaintenance, Knowledge discovery, maintenance decision support.
11 INTRODUCTION
INTRODUCTION aim to support maintenance decision-making. These artefacts
aim to support maintenance decision-making. These artefacts
also enable improvement of different maintenance approaches,
1The also enable improvement of different maintenance approaches,
The dynamic
INTRODUCTION
dynamic global
global and
and local
local business
business scenarios
scenarios put
aim put new
suchtoas;support
new
preventive maintenance
maintenance decision-making.
and correctiveThese artefacts
maintenance.
demands on the decision-making processes in an such as;
organisation.
also enable preventive
improvement maintenance
of and
different corrective maintenance.
maintenance approaches,
demands
The dynamicon theglobal
decision-making
and local processesscenarios
business in an organisation.
Furthermore, ICT provide additional capabilities, which can be
put new
The Furthermore, ICT provide additional andcapabilities, which can be
The new
new decision-making
demands decision-making
on the
processes need
processes
decision-making
need to
processes
to provide
provide enhanced
in an
suchenhanced
used as; preventive
within
organisation.
maintenance
diagnostic and prognostic corrective maintenance.
processes. The
capability forfor knowledge
knowledge discovery
discovery online
online andand in used
in real-time.within ICT
real-time. To
Furthermore,To diagnostic
provide and prognostic
additional capabilities, processes.
whichcan canThe
be
capability
The new the decision-making processes needorganisations
to provide prognostic and diagnostic processes in an enterprise
enhanced be
increase overall business efficiency, prognostic
used need to
within and diagnostic
diagnostic processes
and in an enterprise
prognostic processes. canThebe
increase
capability the
foraoverall
knowledgebusiness efficiency,
discovery online organisations
and ininreal-time.needToto
facilitated through provision of proper information logistics
implement knowledge discovery platform facilitated
their
prognosticcore through provision of proper information logistics
implement
increase aoverall
knowledge discovery platform aimed to and
in their core supportdiagnostic processes
maintenance in an
decision enterprise
making through can be
processesthesuch as business
business,efficiency,
operation, organisations aimed
and maintenance.need
facilitated
to
to through
support provision
maintenance of decision
proper making logistics
information through
processes
implement such
a as business,
knowledge operation,
discovery platformand in maintenance.
provision ICT-based solution for data processing and
their core
Knowledge discovery is depended on availability if accurate provision
aimed to ICT-based
support solution for
maintenance data making
decision processing and
through
Knowledge
processes discovery
such as and is depended
business, on availability
operation, if accurate
information
and maintenance. management (Karim, 2008).
and consistence data information. information management (Karim, 2008).
provision ICT-based solution for data processing and
and consistence
Knowledge data and
discovery is information.
depended on availability if accurate
In the context
information of complex
management (Karim, technical systems information
2008). systems
Today,
and enterprises
consistence are overwhelmed
data In the context
by managing data and its of complex technical information
Today, enterprises areand information.by
overwhelmed managing data and its
logistics refers to: 1) time management, which addresses when
logistics. It can be testified that there is a growing gap between logistics
In refers to:of1) complex
time management, which addresses when
logistics.
Today, It can be testified
enterprises are that there is
overwhelmed byamanaging
growing gap to deliver; 2) content management, which refers information
the context
between
data2011).
and its
technical systems to what to
data generation and data understanding (Witten et. al, to deliver;
logistics It 2) to:
content management, which refers to what to
data generation
logistics. It can andtestified
be data understanding
that there is (Witten
a growing deliver; 3) communication management, which refers towhen
et.gap
al, 2011). refers
between It 1) time management, which addresses how
can be considered that decisions are also becoming more deliver;
to 3) communication management, which refers to how
can
data be considered
generation that understanding
and data decisions are (Witten
also becoming
et. to deliver;
deliver; 2)
more
al, pressure,
2011). It 4) content
context management,
management, which which refers
addresses to what
whereto
complex with greater uncertainty, increasing timeto deliver;
deliver; 3) 4) context management,
communication management, which addresses
which refers where
to how
complex
can be with greater
considered thatuncertainty,
decisions increasing
are also time pressure,
and why to deliver (Karim, 2008; Karim et al, 2009; Kajko-
becoming more
more rapidly changing conditions, and higher stakes and
to why to deliver
deliver; 4) 2011).
context (Karim, 2008; Karim
management, which et addresses
al, 2009; Kajko-
where
more
complex rapidly changing
with&greater conditions,
uncertainty, and higher
increasing timeMattson stakes
et al.,
pressure, However, ICT-based solutions to develop
(Busemeyer Pleskac, 2009). The increased Mattson ettoal.,
information
and why 2011).(Karim,
deliver However, ICT-based
2008; Karim solutions
et al, to develop
2009; Kajko-
(Busemeyer
more rapidly & changing
Pleskac, 2009). The increased
conditions, and information
and establish an effective and efficient information logistics for
higher stakes
needs and the development of Information and Communication and establish
Mattson et al.,an effective and efficient information logistics for
needs and the &
(Busemeyer development
Pleskac, of Information
2009). The and Communication
increased prognostics
information and2011). However,
diagnostics in anICT-based
enterprisesolutions to develop
can be materialised
Technology (ICT) have added velocity to everything that is prognostics
and establish and
an diagnostics
effective andinefficient
an enterprise can be materialised
information logistics for
Technology
needs and thean (ICT) have added
development velocity toand everything that is
through eMaintenance solution (Sderholm et al, 2009). The
done within organizationofthrough
Information
transforming Communication
through eMaintenance
the business
prognostics and diagnosticssolution
in an (Sderholm
enterprise canet be
al,materialised
2009). The
done within
Technology an organization
(ICT) have(Lee, addedthrough transforming
velocity to everythingthe business
provision of the right information to the right information
that is
process into eBusiness 2003). The knowledge provision
discovery,
through of the right information to the right al, information
process
done into an
within eBusiness (Lee, through
organization 2003). The consumereMaintenance
knowledgethediscovery,
transforming business and producersolution with the(Sderholm
right qualityetand 2009). The
at the right
which is an essentially a major aspect for maintenance decision consumer
provision and
of producer
the right with the
information rightto quality
the and
right at the right
information
which
process is an essentially a major aspectThefor knowledge
maintenance decision
time is essential (Parida et al., 2004; Parida, 2006). Therefore,
support;into eBusiness
is usually done(Lee, 2003).
by discovering special pattern discovery,
time is essentialproducer
of data,
consumer (Parida withet al.,the
2004;
rightParida, 2006).
and atTherefore,
support;
which is is
an usually donea by
essentially majordiscovering
aspect forspecial
maintenance there is a and
patterndecision
of data, need to integrate knowledge quality
discovery the right
with the
i.e. by clustering together data that share certain common there
time isessential
is a need (Parida
to integrate
et al., knowledge
2004; Parida, discovery
2006). with the
Therefore,
i.e. by clustering
support; together data that sharespecialcertain
patterncommon
maintenance decision support for effective decision making.
propertiesis (Wang,
usually 1997).
done by discovering of data,
maintenance
there is a need decision support for
to integrate effective discovery
knowledge decision making.with the
properties (Wang, 1997).
i.e. by clustering together data that share certain common Decision-making
maintenance decision in maintenance
support for has to decision
effective be augmentedmaking. to
Extensive application
properties (Wang, 1997). of ICT and other emerging technologies Decision-making in maintenance has to be augmented to
Extensive application of ICT and other emerging technologies instantly understand and efficiently act, i.e. the new know. The
facilitate easy and effective collection of data and information instantly understand
Decision-making in and efficiently act,
maintenance i.e.
to the
be new know. The
facilitate
Extensive easy and effective
application collection ofemerging
data andtechnologies
information
new know in maintenance needs tohas focus on augmented
two aspects of to
(Parida, 2006; CandellofetICT and other
al., 2009). In maintenance, new know
enhanced
instantly in maintenance
understand and needs
efficiently to focus
act, i.e. on
the two
new aspectsThe
know. of
(Parida,
facilitate 2006;
easy Candell
and et al.,collection
effective 2009). In of maintenance,
data and enhanced
knowing: 1) what can be known and 2) what must be known,
information
use of ICT facilitates the development of artefacts knowing:
new know 1)inwhat
(e.g. can be known
maintenance needsand to 2) what
focus onmust
two be known,
aspects of
use of 2006;
(Parida, ICT facilitates
Candell et the development
al., 2009). of artefacts
In maintenance, in order (e.g.
to enable
enhanced the maintenance decision-makers to perform
frameworks, tools, methodologies and technologies);in order to1)
knowing: which enable
what the
can maintenance
be known decision-makers
and 2) what must to
be perform
known,
frameworks,
use of ICT tools, methodologies
facilitates and technologies);
the development of artefactswhich
appropriate actions.
(e.g.
appropriate actions.
in order to enable the maintenance decision-makers to perform
frameworks, tools, methodologies and technologies); which
appropriate actions.
2405-8963 2016, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.
Copyright 2016 IFAC 214
Copyright
Peer reviewunder
2016 responsibility
IFAC 214Control.
of International Federation of Automatic
10.1016/j.ifacol.2016.11.037
Copyright 2016 IFAC 214
2016 IFAC AMEST

October 19-21, 2016. Biarritz, France Ramin Karim et al. / IFAC-PapersOnLine 49-28 (2016) 214219 215

The recent emerging technological and methodological performance. In this context, a cost driver is an input element
achievements in knowledge discovery and knowing are to a process that causally affects or drives costs which is
changing the knowledge industry. Subsequently, the tangible. A cost driver can be interpreted as an element that
maintenance as a part of the new knowledge industry needs to affects cost, or an element that increases costs considerably.
be adapted to this new reality. The ongoing industrial Examples: Assets with high failure frequency and long
digitalization provides enormous capabilities for industry to downtime.
collect vast amount of data and information (i.e. Industrial Big
Some of the key parameters in the form of performance
Data), from various processes and data sources such as
measures or indicators for Reliability Availability
operation, maintenance, and business processes. However,
Maintainability and Supportability (RAMS) and Life Cycle
having accurate data and information available is one the
Cost (LCC), etc. are continuously developed and applied for
prerequisites in maintenance knowledge discovery. Beside the
tracking maintenance activities.
collecting data and information, another puzzle in the new
maintenance knowing is to understand the patterns and Maintenance decision making, with multiple stakeholders
relationships of these data collections. largely depends on the maintenance failure data, RAMS and
LCC data analysis for estimating the Remaining Useful Life
Hence, the purpose of this paper is to propose a concept for
(RUL) in order to perform effective business and maintenance
knowledge discovery in maintenance with focus on Big Data
decision making. Thus, supporting an effective maintenance
and analytics. The concept is called Maintenance Analytics
decision making process needs a trusted DSS based on
(MA). MA focuses in the new knowledge discovery in
knowledge discovery. The process of knowledge discovery will
maintenance. MA addresses the process of discovery,
essentially consists of; data acquisition, to obtain relevant data
understanding, and communication of maintenance data from
and manage its content; data transition, to communicate the
four time-related perspectives, i.e. 1) Maintenance
collected data; data fusion, to compile data and information
Descriptive Analytics (monitoring) focuses to discover and
from different sources; data mining, to analyse data to extract
describe what happened in the past; 2) Maintenance
information and knowledge; and information extraction and
Diagnostic Analytics focuses to understand why something
visualization, to support maintenance decision; as shown in
happened; 3) Maintenance Predictive Analytics focuses to
Figure 1.
estimate what will happen in the future; and 4) Maintenance
Prescriptive analytics which addresses what need to be done
next.
2 MAINTENANCE DECISION-MAKING AND
KNOWLEDGE DISCOVERY
Assets are complex mixes of complex systems, built from
components which, over time, may fail. The ability to quickly
and efficiently determine the cause of failures and propose
optimum maintenance decisions, while minimizing the need for
Figure 1. A generic knowledge discovery process.
human intervention is necessary. Thus, for complex assets,
much information needs to be captured and mined to assess the The integration of data, recorded from a multiple-sensor
overall condition of the whole system. Therefore, the system, together with information from other sources to
integration of asset information is required to get an accurate achieve inferences is known as data fusion (Hall & Llinas,
health assessment of the whole system, and determine the 2001). Data fusion is a prerequisite when handling data from
probability of a shutdown or slowdown. Moreover, the data heterogeneous sources or from multiple sensors. Knowledge
collected are not only huge but often dispersed across discovery when applied for maintenance decision support uses
independent systems that are difficult to access, fuse and mine eMaintenance concept for integrating the data mining and
due to disparate nature and granularity. Data relevant to asset knowledge discovery. However, development of eMaintenance
management are gathered, produced and processed on different for industrial application faces a number of challenges which
levels by different IT systems (Galar et al., 2012; Kans, 2013; can be categorised into: 1) Organisational; 2) Architectural; 3)
Zhang and Karim, 2014) e.g. ERP (Enterprise Resource Infra-structural; 4) content and contextual; and 5) integration.
Planning) for business functions; SCADA (Supervisory
Organisational challenges mainly focus on aspects related
Control and Data Acquisition) for monitoring process
enterprise resource management. Examples of these challenges
Nowadays the challenge is to provide intelligent tools to
are: 1) restructuring of the organizations involved in
monitor and manage assets (machines, plants, products, etc.)
eMaintenance; 2) planning of resources (e.g. material, spare-
proactively through ICT, focusing on health degradation
part); 3) information management; 4) knowledge management;
monitoring and prognosis instead of fault detection and
and 5) management of heterogeneous organizations.
diagnostics. Maintenance effectiveness depends on the quality,
timeliness, accuracy and completeness of information related Architectural challenges deal with issues related to the overall
to machine degradation state. This translates into following key architecture of eMaintenance solutions. Some of these
requirements: preventing data overload, ability to differentiate challenges are: 1) development of a framework for
and prioritize data (during collection as well as. Tools and development of eMaintenance; 2) development of models for
machinery with satisfactory performance; efficient analysis decentralized data processing and analysis; 3) development of
tools, and planning tools. A performance killer is an input service model for decentralized data analysis; 4) development
element to a process that performs poorly or hinders

215
2016 IFAC AMEST
216
October 19-21, 2016. Biarritz, France Ramin Karim et al. / IFAC-PapersOnLine 49-28 (2016) 214219
of model based prognostic tools; 5) development of model
aimed for data and information visualization to support
Human-Machine Interaction; and 6) development of model
aimed for distributed data storage capability.
Infra-structural challenges address with issues related to
provision of necessary technologies and tools that are required
to meet needs and requirements when services, according to
SOA, are developed, implemented and managed in an
enterprise. Example of these are: 1) network infra-structure
(e.g. wired and wireless); 2) authentication of services and
users; 3) authorisation of services and users; 4) safety and
security mechanism; 5) maintainability of eMaintenance
services; 6) availability performance management; and tracing
and tracking mechanism.; and 7) provision of mechanism
aimed for documentation and archiving
Content and contextual challenges are mainly related to data
and information provided through the eMaintenance services.
Some of these challenges are: 1) provision of appropriate
ontology through which data from data sources (e.g. process
data, product data, condition monitoring data, and business
data) can smoothly and seamlessly be integrated; 2) provision
of quality assurance mechanism that ensures that required data
quality is fulfilled and visualised, in order to increase the
quality of decision making; 3) mechanism for sensing users
current situation in order to adapt information to users context;
4) provision of mechanism for describing various context; 5)
mechanism to manage uncertainty in data sets; and 6) provision
of mechanism for pattern recognition.
2.1 Data Mining and Knowledge Discovery
Data produced in asset management can be described in terms
of the 5Vs de-scribed by Zikopoulos and Eaton (2011) and
(Lomotey and Deters (2014). Data from sensors like
accelerometers or acoustic sensors can be acquired with
velocity of tens of thousands of samples per second per each
measuring point. Having hundreds or thousands of those points,
big volume of data is being produced. Some maintenance
related data are structured while some are not, such as free text
comments for performed maintenance actions or failure
reports. Moreover, data from different systems are in different
formats. This is the source of variety of data in asset
management. Those data has potential value when properly
employed in asset management, but in order to achieve this,
there is need to asses and man-age the veracity of the data, i.e.
the data uncertainty. Finally, understanding the value of data,
i.e. how data can enable efficiency and effectiveness in
maintenance management, for instance for improved decision
making, and to choose the most cost-effective means to process
the data is important.
Data mining in big asset data can discover knowledge in terms
of new patterns and relations not visible at a glance. The big
data approach enables incorporation of contextual information
in Maintenance Decision Support Systems (Galar et al., 2015).
One example of useful knowledge that could be discovered is
root causes of failure. This can provide an input for design
improvement, as well as for more accurate maintenance
planning. Indeed, more and more authors are using the term
KDD synonymously today. They refer to the area of research
that draws upon data mining methods from pattern recognition
(Tuzhilin, 1993), machine learning (Han et al., 1992) and
216
database (Agrawal et al., 1993, 1994) techniques in the context
of vast organizational databases. Conceptually, KDD refers to
a multiple step process that can be highly interactive and
iterative in the following (Fayyad & Uthurusamy, 1995; Wang,
1997):
the selection, cleaning, transformation and projection of
data;
mining the data to extract patterns and appropriate
models;
evaluating and interpreting the extracted patterns to
decide what constitutes knowledge;
consolidating the knowledge, resolving conflicts with
previously extracted knowledge;
making the knowledge available for use by the interested
elements within the system.
Artificial intelligent techniques have advanced knowledge
management, including knowledge acquisition, knowledge
repositories, knowledge discovery, and knowledge
distribution. Knowledge acquisition captures tacit and explicit
knowledge from domain experts, while knowledge repositories
formalize the outcomes of knowledge acquisition and integrate
knowledge in distributed corporate environments. Knowledge
discovery and mining approaches explore relationships and
trends in the knowledge repositories to create new knowledge
(Liu and Ke, 2007).
3 THE STUDIED CASE
3.1 Railway Cloud General Description
The railway cloud, or E365 Analytics, is a set of decision
support services aimed to achieve business excellence in
industry. E365 Analytics is built on edge technologies for
information & communication provisioning. E365 Analytics
provides services for the whole decision-making process, such
as data acquisition, data transformation, data fusion &
integration, data processing, Big Data analysis, context
adaptation, and information visualisation. The conceptual
model is depicted in Figure 2. E365 Analytics provides a set
of interconnected, loosely coupled services, which can be
orchestrated to fulfil customers demands on decision support.
These services are built upon technologies such as Big Data,
text analytics, data analytics, context awareness, sensor fusion,
cloud computing, and crowd-computing.
Figure 2. The conceptual model of E365 Analytics.
The conceptual model of E365 Analytics consists of five
steps and is similar to the generic information process
presented in Figure 2.
3.2 Railway Cloud Swedish railway data.
E365 Analytics do not generate any system-related technical
data itself so in order to deliver maintenance decision support
in the Swedish railway system it is dependent on the acquisition
2016 IFAC AMEST
Ramin Karim et al. / IFAC-PapersOnLine 49-28 (2016) 214219 217
October 19-21, 2016. Biarritz, France

of relevant data and a automation of that process. E365 To enable this flexibility a modular architecture is used in the
Analytics is built to acquire and manage all types of web interface, the modular architecture is depicted in Figure 3.
maintenance data, business data and condition data for the
Swedish railway.
The business and maintenance data structure and the related
services to collect these data are mostly customer specific. This
due to varying character on customers but also on the large
numbers of stakeholders involved and IT strategies, among
other things. This applies to most of the condition data as well
but there exist some shared data channels.
The Swedish Transport Administration (Trafikverket), which
is the infrastructure owner offers some condition monitoring to
all operators on the rail. They have equipped the Swedish
railway network with hundreds of stationary detectors and
measurement stations. These detectors monitors, and alarms, Figure 3. Modular web framework architecture.
passing trains according to a number of parameters including
By using this architecture small apps, with simpler
box and wheel temperatures, forces and wheel profiles.
functionality, accessing only the relevant data for the intended
Operators having installed RFID tags, consisting of the
user can be developed. A set of these apps, suiting the
European Vehicle Number (EVN), on their vehicles may
customers needs, can then be selected for each customer.
acquire this condition data from Trafikverket.
3.4 Railway Cloud Usage
In addition to condition data from stationary detectors, mobile
systems can be included as well. This can be anything form From a user perspective the interaction with the data is done
detachable systems monitoring the infrastructure, to a train on- through a web platform including a set of railway apps. The
board systems delivering information about its vehicles. railway apps can be divided into three main categories, Rolling
Stock, Infrastructure and Tools. Rolling stock focuses on the
Considering the large number of different types of data in each vehicle as the maintenance entity and enables the user to
class, the various interactions needed to collect data from monitor the condition on all vehicles as a population, a subset
customers, the span width in data amount which can vary from of vehicles or an individual vehicle. Below in Figure 4 the
a small number of data flows to a Big Data problem, it is clear Vehicle Set app is presented in which the user can monitor a
that the most collection and handling of data services must be subset of vehicles in reference to the full population of vehicles
customer specific. This puts requirements on a railway cloud to in the same class.
be highly scalable but also flexible in the interaction with
customers and their users.
3.3 Railway Cloud Service architecture
In Figure 2 it is depicted that E365 Analytics is built to fill
the gap between data generation and decision making. The
process of transforming the generated raw data to relevant
information to the decision maker can, on a higher level, be
divided into five steps. The first step, data acquisition, refers to
the process of retrieving and structuring data. Data collection
can be done from all types of sources which gives the customer Figure 4. A subset of vehicles compared to the full set.
the opportunity to include all relevant data of any format. Data
processing is the step of cleaning and validating the collected Infrastructure apps target users more interested in stationary
data. This is an important step regarding quality and requires assets. Instead of focusing up from the rail the focus is moved
knowledge about the information models and the capacity to down to the rail. Assets in interest may be rail, detectors,
treat data sources as individuals. measurement stations, sidings or other closed related entities.
In data integration multiple sources or multiple data types can The Tools category exists in order not to limit the user in terms
be integrated. By integrating multiple sources of the same data of analysis. With Tools apps the user can compare
types it is possible to compare parameters and by that retrieve distributions, see seasonal changes and analyze time series of
an indication on the quality of the data, integrating multiple data. Figure 5 gives an example how the Time Series Analysis
data types enables a more thorough analysis utilizing the tool can be used to compare flange height for all four wheels
relation between the data types and their behaviour. The last on one side of a vehicle.
step, Data visualization, is the interface between the user and
the cloud. This includes any interaction from the user. As
previously mentioned, this turns into quite a complex task due
to the varying character of the customers and their users. The
information to visualize and the relevancy in the information
will differ a lot depending on customer characteristics and user
level of operation.

217
2016 IFAC AMEST
218 Ramin Karim et al. / IFAC-PapersOnLine 49-28 (2016) 214219
October 19-21, 2016. Biarritz, France
Figure 5. Description of meta-level knowledge provisioning.
4 THE MAINTENANCE ANALYTICS CONCEPT
To support maintenance knowing and decision-making
smoothly and efficiently in an era, when the technological
revolution such as IoT internet-of-things, industrial internet,
industry 4.0, and smart factories enables an enhanced
availability of data and information, there is a need of a
structured approach and concept to improve information
extraction and knowledge discovery. For this purpose a concept
has been proposed to which aims to deal with analytics in
maintenance from methodological and technological
perspectives.
The proposed concept for Maintenance Analytics (MA) is
based on four interconnected time-lined phases, which aim to
facilitate maintenance actions through enhanced understanding
of data and information. The MA phases are (see Figure 6): 1)
Maintenance Descriptive Analytics; 2) Maintenance
Diagnostic Analytics; 3) Maintenance Predictive Analytics;
and 4) Maintenance Prescriptive analytics.
Figure 6. The constitution phases of Maintenance Analytics.
In maintenance decision-making the process of dealing with
fault, i.e. inability to perform as required, due to an
internal state (IEV, 2016), and failure, i.e. the termination of
the ability of an item to perform a required function (IEV,
2016) is essential. The relationship between fault and failure
can be described as a fault is result (state) of an associated
failure (event). However, in order to deal with the event and
state in a system condition, the proposed MA phases have been
ordered in a time-related order.
4.1 Maintenance Descriptive Analytics
The Maintenance Descriptive Analytics phase of MA aims to
answer What has happened?. In this phase access to data
related to system operation, system condition, and expected
condition is highly important. Another important aspect in
order understand the relationship of events and states during the
descriptive analytics is time and time frame associated with
each specific log. Furthermore, events and states need to be
218
associated with the system configuration for the time (T n). This
means that time synchronisation becomes an important part to
support maintenance analytics.
4.2 Maintenance Diagnostic Analytics
The Maintenance Diagnostic Analytics phase of MA aims to
answer Why something has happened?. In this phase the
outcome from Maintenance Descriptive Analytics is used to
frame the analytics. For this phase, availability of reliability
data is necessary beside the data used in descriptive phase.
4.3 Maintenance Predictive Analytics
The Maintenance Predictive Analytics phase of MA aims to
answer What will happen in the future?. In this phase the
outcome from Maintenance Descriptive Analytics is used.
Additionally, in this phase, availability of reliability data and
maintainability data is necessary beside the data used in
descriptive phase. In addition, in order to predict upcoming
failure and fault there is a need to provide business data such as
planned operation and planned maintenance to this phase.
4.4 Maintenance Prescriptive analytics
The Maintenance Prescriptive Analytics phase of MA aims to
answer What needs to be done?.
In this phase the outcome from Maintenance Diagnostic
Analytics and Maintenance Predictive Analytics are used. In
addition, in order to predict upcoming failure and fault there is
a need to provide resource planning data and business data.
5 Maintenance Analytics and eMaintenance
When dealing with Maintenance Analytics (MA) provision of
appropriate information logistics is essential. The main aim of
information logistics is to provide just-in-time information to
targeted users and optimization of the information supply
process, i.e. making the right information available at the right
time and at the right point of location (Heuwinkel et al., 2003;
Haseloff, 2005). Solutions for in-formation logistics need to
deal with: I) time management, which addresses when to
deliver; II) content management, which refers to what to
deliver; III) communication management, which refers to
how to deliver; IV) con-text management, which addresses
where and why to deliver (Heuwinkel et al., 2003; Haseloff,
2005). In maintenance, eMaintenance is defined as
materialisation of information logistics aimed to support
maintenance decision-making (Karim, 2008; Kajko-Mattson et
al, 2011). In order to develop eMaintenance solutions for
supporting the MA concept, there is a need of overarching
approach to combine modelling of data, knowledge, and
context, as illustrated in Figure 9. Furthermore, when dealing
with information integration, there are essential issues that need
to be considered: I) the syntactic problem; which refers to the
format and structure of content, and II) the semantic problem;
which refers to the meaning of content. Further, for an
application area, e.g. maintenance, it is also necessary to apply
an ontology, which is a representational vocabulary for a shared
domain that includes definitions of classes, relations, functions,
and other objects (Gruber, 1993). Ontology is also described as
an explicit specification of a conceptualization (i.e. abstract,
simplified view of the world), which is the objects, concepts,
and other entities that are presumed to exist in some area of
interest and the relationships that hold them (Gene-sereth &
2016 IFAC AMEST
October 19-21, 2016. Biarritz, France Ramin Karim et al. / IFAC-PapersOnLine 49-28 (2016) 214219 219
Nilsson, 1987). Furthermore, ontology is described as the
representation of the meaning of terms in vocabularies and their
interrelationships (W3C, 2008). A common ontology facilitates
the machine processability and supports reasoning outside the
transaction context.
Figure 7. eMaintenance solution for Maintenance Analytics.
However, eMaintenance implies a wide range tools,
technologies, and methodologies aimed for maintenance
decision-making, including analytics. Hence, eMaintenance as
be considered as a concept through which MA can be
materialised, as illustrated in Figure 7.
6 Maintenance Analytics and Big Data
As described in above, the inherent phases of Maintenance
Analytics are highly dependent on availability of vast amount
data from various data sources. Big Data is term used to
illustrated complex datasets with respect to: 1) Volume; 2)
Velocity; 3) Variety; and 4) Veracity. Whereas, analytics is the
process of generate knowing based on understanding.
However, MA is considered as concept for big data analytics
for maintenance. From a technological perspective MA focuses
on some key aspects such as: cloudification; service-
orientation; process-orientation; distributed computing;
modularisation, context adaptation; and usability.
7 CONCLUSIONS
The purpose of this paper was to propose a concept for
knowledge discovery in maintenance with focus on big data
and analytics. In the new concept for big data analysis in
maintenance has been provided, called Maintenance Analytics
(MA). MA consists of four interconnected phases, which are:
1) Maintenance Descriptive Analytics (monitoring); 2)
Maintenance Diagnostic Analytics; 3) Maintenance
Predictive Analytics; and 4) Maintenance Prescriptive
analytics. In order to validate the proposed concept a major case
study has been conducted.It can be concluded that the demands
on maintenance decision-making and knowledge discovery has
dramatically been changed due to technological revolution.
This has meant that the amount of maintenance-related data and
information has also increased dramatically. Furthermore, it
can be concluded that, in order to satisfy the business needs
with respect decision-making, there is need of a systematic
approach and a corresponding concept to deal with knowledge
discovery in maintenance. One concept which provides a
systematic approach to maintenance knowledge discovery has
been introduced in this, i.e. MA. The conducted case study
shows that MA materialised through edge technology can
facilitate the decision-making process in maintenance. The
study also shows that from a scientific perspective the
streamlining of analytics in maintenance through e.g. proposed
219
concept for maintenance analytics supports selection of
appropriate scientific methodologies and technologies.
ACKNOWLEDGEMENTS
Hereby, we want express our appreciation to eMaintenance365
AB, The Swedish Rail Administration (Trafikverket), LKAB
(the Swedish mining company), and the ePilot project for
funding and undertaking the research study.
REFERENCES
Agrawal, R., Carey, M., & Faloutsos, C. (1994). Quest: A project on database
mining. Sigmod Record, 23(2), 514
Agrawal, R., Imielinski, T., & Swami, A. (1993). Database mining: A
performance perspective. IEEE Transactions on Knowledge and Data
Engineering, 5(6), 914-925.
Bolloju, N., Khalifa, M. & Turban. E. (2002). Integrating knowledge
management into enterprise environments for the next generation decision
support. Decision Support Systems, 33, 163 - 176
Busemeyer, J. R., & Pleskac, T. J. (2009). Theoretical tools for understanding
and aiding dynamic decision making. Journal of Mathematical Psychology,
53(3), 126-138.
Candell O., Karim R. & Sderholm P. (2009). eMaintenanceInformation
logistics for maintenance support. Journal of Robotics and Computer-
Integrated Manufacturing, 25(6), pp. 937944.
Fayyad, U. M., & Uthumsamy, R. (1995). Preface. In Proceedings of the First
International Conference on Knowledge Discovery and Data Mining. New
York: AAAI Press.
Hall, D. & J Llinas, J. (2001). Handbook of multisensor data fusion, pp 17-
26, CRC Press LLC, Boca Raton, USA.
Han J., Cai Y., Cercone N., & Huang, Y. (1992). DBLEARN: A knowledge
discovery system for large databases. International Conference on
Information and Knowledge Management, Baltimore, MD.
IEV (2016). International Electrotechnical Vocabulary (IEV). Available at:
http://www.electropedia.org/, accessed: 18 March 2016.
Kajko-Mattsson, M., Karim, R., & Mirjamdotter, A. (2011). Essential
Components of eMaintenance. International Journal of Performability
Engineering, 7(6), November 2011, pp. 505-521.
Karim, R. (2008). A Service-Oriented Approach to eMaintenance of Complex
Technical Systems. Doctoral thesis, Lule University of Technology, Lule,
Sweden.
Karim, R., Candell, O. and Sderholm, P. (2009). Development of ICT-based
maintenance support services. In Journal of Quality in Maintenance
Engineering, vol. 15, nr. 2, pp. 127 150
Lee, J. (2003). E-manufacturing--fundamental, tools, and transformation.
Robotics and Computer-Integrated Manufacturing, 19 (6), pp. 501-507.
Liu, D, & Ke, C. (2007). Knowledge support for problem-solving in a
production process: A hybrid of knowledge discovery and case based
reasoning. Expert Systems with Applications. 33, 147-161
Parida, A. (2006). Maintenance performance measurement system:
Application of ICT and e-Maintenance Concepts, International journal of
COMADEM, Vol.9, No. 4, 30-34
Parida, A., Phanse, K. and Kumar, U. (2004).An integrated approach to
design and development of e-maintenance system. Proceedings of
VETOMAC-3 and ACSIM-2004, New-Delhi, 6-9 December 2004.
Sderholm, P., Karim, R., Candell, O. & Kumar, U. (2009). In Proceedings of
CM/MFPT 2009 - (CM2009), held 23-25 June 2009, Dublin, Ireland.
British Institute of Non-Destructive Testing and Coxmoor.
Tsang, A. H. C. (2002). Strategic dimensions of maintenance management,
Journal of Quality in Maintenance Engineering, Vol. 8, No. 1, pp. 7-39.
Tseng, S,-F. (1997) Diverse reasoning in automated model formulation.
Decision support systems, 20 (4), 357 383
Turban, E; and Aronson, J. (1998). Decision support systems and Intelligent
systems, Prentice Hall, Upper Saddle River, WJ.
Tuzhilin, A. (1993). Abstract-driven pattern discovery in databases. IEEE
Transactions on Knowledge and Data Engineering, 5(5). 926-938.
Wang, H. (1997). Intelligent agent-assisted decision support systems:
integration of knowledge discovery, knowledge analysis, and group
decision support. Vol. 12, No.3, 323-335
Witten, I.H., Frank, E., and Hall, M A (2011). Data Mining Practical
Machine Learning Tools and Techniques. Morgan Kaufmann Elsevier.