Content uploaded by Alexandros Bousdekis
Author content
All content in this area was uploaded by Alexandros Bousdekis on Jul 27, 2016
Content may be subject to copyright.
Industrial Management & Data Systems
A proactive decision making framework for condition-based maintenance
Alexandros Bousdekis Babis Magoutas Dimitris Apostolou Gregoris Mentzas
Article information:
To cite this document:
Alexandros Bousdekis Babis Magoutas Dimitris Apostolou Gregoris Mentzas , (2015),"A proactive
decision making framework for condition-based maintenance", Industrial Management & Data
Systems, Vol. 115 Iss 7 pp. 1225 - 1250
Permanent link to this document:
http://dx.doi.org/10.1108/IMDS-03-2015-0071
Downloaded on: 29 July 2015, At: 04:29 (PT)
References: this document contains references to 62 other documents.
To copy this document: permissions@emeraldinsight.com
The fulltext of this document has been downloaded 30 times since 2015*
Users who downloaded this article also downloaded:
Farhad Panaihfar, Cathal Heavey, PJ Byrne, (2015),"Developing retailer selection factors for
collaborative planning, forecasting and replenishment", Industrial Management & Data Systems,
Vol. 115 Iss 7 pp. 1292-1324 http://dx.doi.org/10.1108/IMDS-01-2015-0009
Access to this document was granted through an Emerald subscription provided by emerald-
srm:199354 []
For Authors
If you would like to write for this, or any other Emerald publication, then please use our Emerald
for Authors service information about how to choose which publication to write for and submission
guidelines are available for all. Please visit www.emeraldinsight.com/authors for more information.
About Emerald www.emeraldinsight.com
Emerald is a global publisher linking research and practice to the benefit of society. The company
manages a portfolio of more than 290 journals and over 2,350 books and book series volumes, as
well as providing an extensive range of online products and additional customer resources and
services.
Emerald is both COUNTER 4 and TRANSFER compliant. The organization is a partner of the
Committee on Publication Ethics (COPE) and also works with Portico and the LOCKSS initiative for
digital archive preservation.
*Related content and download information correct at time of
download.
Downloaded by NATIONAL TECHNICAL UNIVERSITY OF ATHENS At 04:29 29 July 2015 (PT)
A proactive decision making
framework for condition-based
maintenance
Alexandros Bousdekis and Babis Magoutas
Information Management Unit, Institute of Communications and
Computer Systems (ICCS), National Technical University of Athens,
Athens, Greece
Dimitris Apostolou
Department of Informatics, University of Piraeus, Piraeus, Greece, and
Gregoris Mentzas
Information Management Unit, Institute of Communications and
Computer Systems, National Technical University of Athens, Athens, Greece
Abstract
Purpose –The purpose of this paper is to perform an extensive literature review in the area of
decision making for condition-based maintenance (CBM) and identify possibilities for proactive online
recommendations by considering real-time sensor data. Based on these, the paper aims at proposing a
framework for proactive decision making in the context of CBM.
Design/methodology/approach –Starting with the manufacturing challenges and the main
principles of maintenance, the paper reviews the main frameworks and concepts regarding CBM that
have been proposed in the literature. Moreover, the terms of e-maintenance, proactivity and decision
making are analysed and their potential relevance to CBM is identified. Then, an extensive literature
review of methods and techniques for the various steps of CBM is provided, especially for prognosis
and decision support. Based on these, limitations and gaps are identified and a framework for proactive
decision making in the context of CBM is proposed.
Findings –In the proposed framework for proactive decision making, the CBM concept is enriched in the
sense that it is structured into two components: the information space and the decision space. Moreover, it
is extended in a way that decision space is further analyzed according to the types of recommendations
that can be provided. Moreover, possible inputs and outputs of each step are identified.
Practical implications –The paper provides a framework for CBM representing the steps that need
to be followed for proactive recommendations as well as the types of recommendations that can
be given. The framework can be used by maintenance management of a company in order to conduct
CBM by utilizing real-time sensor data depending on the type of decision required.
Originality/value –The results of the work presented in this paper form the basis for the development
and implementation of proactive Decision Support System (DSS) in the context of maintenance.
Keywords Decision making, Condition-based maintenance, E-maintenance, Proactivity,
Real-time data, Recommendations
Paper type Research paper
1. Introduction
In manufacturing, equipment maintenance is a significant contributor to the total
company’s cost, so having an optimal maintenance policy in terms of cost, equipment
downtime and quality is an important efficiency enabler (Waeyenbergh and Pintelon, Industrial Management & Data
Systems
Vol. 115 No. 7, 2015
pp. 1225-1250
© Emerald Group Publishing Limited
0263-5577
DOI 10.1108/IMDS-03-2015-0071
Received9March2015
Revised 19 May 2015
Accepted 3 June 2015
The current issue and full text archive of this journal is available on Emerald Insight at:
www.emeraldinsight.com/0263-5577.htm
This work is partly funded by the European Commission project FP7 STREP ProaSense “The
Proactive Sensing Enterprise”(612329).
1225
Proactive
decision
making
framework
Downloaded by NATIONAL TECHNICAL UNIVERSITY OF ATHENS At 04:29 29 July 2015 (PT)
2004; Garg and Deshmukh, 2006). Maintenance is related to all the processes of
a manufacturing firm and focusses not only on avoiding the equipment breakdown but
also on improving business performance, for example, in terms of productivity,
elimination of malfunctions, etc. Various maintenance policies have been examined in
both the academic and industrial realms and a multitude of maintenance strategies
have been recommended in an effort to develop a holistic approach for maintenance
management, which supports both reactive and proactive support maintenance actions
(Waeyenbergh and Pintelon, 2004).
“Proactivity”in the context of information systems refers to the ability to avoid or
eliminate undesired future events or exploit future opportunities by implementing
prediction and automated decision making technologies (Engel and Etzion, 2011).
Proactivity is leveraged with novel information technologies that enable decision
making and support human actions before a predicted critical event occurs.
Application domains that can take advantage of such technologies include transport,
fraud management and maintenance (Artikis et al., 2014; Magoutas et al., 2014).
In manufacturing, sensors have the capability of measuring a multitude of parameters
frequently and collecting plenty of data. Analysis of Big Data, both historical and real-
time, can facilitate predictions on the basis of which proactive maintenance decision
making can be performed.
E-maintenance is related to the notion of proactivity because it supports the
transmission of the enterprise from “fail and fix”to “predict and prevent”concept while,
at the same time, maintenance is addressed as an enterprise process, integrated with
both internal and external business processes (Macchi et al., 2014), for improving
business performance (Lee et al., 2006; Iung et al., 2009). E-maintenance assumes that
data should be available to all enterprise components and actors with the aid of ICT
at the right time and place in order to make optimal maintenance decisions based on
underlying predictions (Iung et al., 2009).
Generally, the need for a business turning from reactive to proactive is increasing.
Proactive enterprise leads to increased situation awareness capabilities even ahead of
time. This will lead to a new class of enterprise systems, proactive and resilient
enterprises, that will be continuously aware of that what “might happen”in the relevant
business context and optimize their behavior to achieve what “should be the best
action”even during stress and balancing on demanding margins. Proactive enterprise
systems will be able to suggest early on to the decision makers the most appropriate
process adjustments to avoid singular system behavior and optimize its performance
(Magoutas et al., 2014).
Although during the last years there have been some efforts toward increasing the
level of proactivity in maintenance decision making, existing approaches are still under
development and suffer from some limitations. The degree of proactivity is usually
low and decisions are narrowed to recommendations about the maintenance schedule,
i.e., the sequence of maintenance actions, the maintenance strategy or, more rarely, the
optimal time of applying a predefined action. In other words, optimization is done for
one criterion at a time, while recommendations involve a general decision. Moreover,
contributions are not presented as part of a wider framework that can support their
integration in manufacturing processes. In addition, the vast majority of prognostic
models are validated within a laboratory environment by doing experiments and not in
industrial settings. This paper aims to review existing works in maintenance decision
making methods and synthesize a generic framework that can support
the development of proactive decision support systems (DSS) that include
1226
IMDS
115,7
Downloaded by NATIONAL TECHNICAL UNIVERSITY OF ATHENS At 04:29 29 July 2015 (PT)
predictions and proactive actions based on these predictions (Engel et al., 2012).
To the best of our knowledge, there is no a holistic e-maintenance framework for
decision making in CBM providing reactive and, even more importantly, proactive
support based on the degradation state of equipment and the prediction of its evolution,
exploiting large amounts of condition data collected automatically by sensors.
In this paper we aim to fill this gap by providing a Proactive Decision Making
Framework for Condition-Based Maintenance (CBM).
The paper is structured in five sections. Section 2 describes the theoretical
background and the motivation for creating a framework for proactive decision
making in the context of CBM. Section 3 provides a review of methods and techniques
used in the steps of CBM, focussing on prognosis and decision support, while Section 4
presents the framework for maintenance decision making. Section 5 presents a
practical demonstration of the proposed framework. Finally, Section 6 discusses the
added value and practical implications of the proposed framework, while Section 7
concludes the paper and presents our plans for future work.
2. Decision making for maintenance operations
2.1 Types of maintenance
Although there is no absolute agreement in the literature about the classification
of maintenance types, we can broadly distinguish between three categories: breakdown
maintenance which takes places when a failure occurs; time-based preventive maintenance
which sets certain activities when a defined period of time passes; and CBM
which recommends actions according to the current and future health state of the
manufacturing system based upon data gathered through condition monitoring
( Jardine et al., 2006).
Breakdown maintenance is the oldest type of maintenance that fixes equipment
as soon as they need to. Time-based preventive maintenance is the evolution of
breakdown maintenance ( Jardine et al., 2006). Time-based preventive maintenance is
widely used in industry; however, companies are increasingly turning to CBM, with
manufacturing companies considering the use of condition monitoring. Currently, even
large manufacturing companies either do not use sensors for measuring indicators of
equipment degradation at all or, even if they do so, they have not developed a complete
CBM strategy in order to utilize its benefits. However, CBM is becoming essential for
every manufacturing business as products have become more and more complex
thanks to the evolution of technology and thus, quality and reliability have become
issues of high significance ( Jardine et al., 2006; Peng et al., 2010). Consequently, the
costs of time-based preventive maintenance have increased and CBM has started to be
evolved as a novel lever for maintenance management (Jardine et al., 2006). CBM is
tightly linked to the notion of proactivity which is the focus of our study and hence we
examine it in detail below.
2.2 CBM
CBM relies on diagnostic and prognostic models and uses them to support decisions
about the appropriate maintenance actions based on the current health state
of a system and/or its predicted performance and remaining lifetime. CBM can be
applied for supporting decisions either about Corrective And Preventive Actions
(CAPA) or about proactive actions. In the first case, only diagnosis is required
so that the actual condition of the system is identified and if it has been failed,
decisions for repair CAPA are taken. In the second case, prognosis is required
1227
Proactive
decision
making
framework
Downloaded by NATIONAL TECHNICAL UNIVERSITY OF ATHENS At 04:29 29 July 2015 (PT)
as well, so that future condition of the system is predicted and decisions
about proactive maintenance actions are taken ( Jardine et al., 2006; Peng et al., 2010;
Voisin et al., 2010).
Several maintenance frameworks have been proposed in the literature outlining the
steps involved in performing CBM. Lee et al. (2004) describes three core steps: first, data
acquisition, to collect the data; second, data processing, to handle the data; and third,
maintenance decision making, to decide about the optimal maintenance policy.
Peng et al. (2010) focussed on the third step (maintenance decision making), further
detailing it into diagnosis and prognosis. The authors also indicated the need for
historical data and for the development of a model for representing system behavior.
Irigaray et al. (2009) focussed on supporting CBM by storing relevant data and
information and utilizing them so that the most appropriate decisions are drawn
and are updated dynamically by means of a platform based on web services and
a systematic process consisting of four layers: condition monitoring, assessment of the
health state, prognosis and decision making.
Peng et al. (2010) described in detail a maintenance decision support framework
consisting of five main steps: first, feature selection, which is conducted with the aid of
historical data as well as several methods such as principal component analysis,
genetic algorithms and support vector machine (SVM); second, data training (analysis);
third, diagnostics and prognostics, by using real-time data; fourth, reliability and
remaining useful life (RUL) where the result is verified and its precision is assessed
in order to give feedback to steps second and fifth; and fifth, maintenance schedule,
which considers the cost function which is extracted from the relationship between
the maintenance cost, RUL and reliability of the system. Figure 1 depicts this
relationship and shows that while time to failure is approaching zero, reliability is
decreasing (Peng et al., 2010). When time to failure becomes zero, a breakdown of the
equipment occurs. The best time to do maintenance is when the maintenance cost is
minimum and reliability has started to increase significantly.
An important principle of CBM is the P-F curve, which can be used to estimate RUL of
some part of equipment. Figure 2 illustrates how a failure starts and deteriorates to the
point at which it can be detected (the potential failure point “P”). Thereafter, if it is not
Time To Failure 0
Maintenance cost Reliability
Best time for
maintenance
Source: Based on Peng et al. (2010)
Figure 1.
Relationship among
RUL, reliability and
maintenance cost
1228
IMDS
115,7
Downloaded by NATIONAL TECHNICAL UNIVERSITY OF ATHENS At 04:29 29 July 2015 (PT)
detected and no suitable action is taken, it continues to deteriorate –usually at an
accelerating rate –until it reaches the point of functional failure (Point “F”). The amount of
time which elapses between the point where a potential failure occurs and the point where
it deteriorates into a functional failure is known as the P-F interval (Veldman et al., 2011).
This interval can be seen as an opportunity window during which actions can be taken
with the aim to eliminate the anticipated functional failure or mitigate its effect.
Arguably the most generic conceptual framework for proactive maintenance
decision support has been proposed by Voisin et al. (2010). This framework considers
the interactions of prognosis with the whole business environment and represents
the business processes which are integrated with prognosis as shown in Figure 3
(Iung et al., 2009; Voisin et al., 2010). Moreover, it separates the decision support step
Point where a failure
starts to occur
Point where it has failed
(functional failure)
Point where it can be found out
that it is failing (potential failure)
Condition
Time
F
P
Figure 2.
P-F curve
Signal Processing
and Acquisition
Data about
system and
environment
Data Monitoring
and Diagnosis
Prognosis
Decision Support
Company Management
Prognostic Knowledge
by prognostic expert
Preferences and choices
by decision maker
Historical data
Maintenance
Management
Source: Adapted from Voisin et al. (2010)
Figure 3.
The role of diagnosis
and prognosis
in CBM
1229
Proactive
decision
making
framework
Downloaded by NATIONAL TECHNICAL UNIVERSITY OF ATHENS At 04:29 29 July 2015 (PT)
from diagnostics and prognostics by combining and updating two earlier frameworks
(Léger and Morel, 2001; Muller et al., 2008a; Lebold and Thurston, 2001).
In the industrial realm, the open systems architecture for CBM (OSA-CBM)
framework has already been implemented in several industries, such as aerospace
industry within the framework of Integrated Vehicle Health Management (IVHM)
(Lebold and Thurston, 2001; Dunsdon and Harrington, 2008; Benedettini et al., 2009).
The OSA-CBM framework consists of seven sequential layers as shown in Figure 4.
Its goal is to enable the integration of prognostics and equipment health management
information from a variety of sources. OSA-CBM describes the entire process of CBM
starting from the collection of data and ending with the decision making step and the
presentation of the results (Lebold and Thurston, 2001; Dunsdon and Harrington, 2008).
2.3 E-maintenance
E-maintenance refers to the convergence of emerging information and communication
technologies with DSS which take into account the resources, services and management
to enable decision making in a proactive way (Muller et al, 2008a). E-maintenance has
become important in the last years due to the emergence of technologies which are able
to optimize maintenance-related workflows and the integration of business performance,
which enable openness and interoperation of e-maintenance with other components
of e-enterprise (Iung et al., 2009). This support does not include only technologies, but also
operations and processes related to maintenance such as condition monitoring, diagnostics,
prognostics, etc. (Muller et al., 2008a; Muller et al., 2008b; Irigaray et al., 2009; Levrat
and Iung, 2007). E-maintenance is considered not only in production and operation
stages but also as an integral part of the whole lifecycle management. Therefore, apart
from production issues, e-maintenance should also embed eco-efficiency and
product design, disassembly and recycling in a way that consists a useful tool for
business process improvement in the context of maintenance lifecycle management
(Takata et al., 2004; Iung et al., 2009).
Presentation
Decision Support
Prognosis
Diagnosis
Condition Monitoring
Signal Processing
Sensor module
GUI-man/machine interface (display data and information to user)
Prediction of future health based on historical and real-time data, diagnosis and knowledge
Automated decision making based on predictions and domain knowledge
Health Assessment with the use of historical and Condition Monitoring data
Comparison of Signal Processing data with pre-defined features
Pre-processing and feature extraction (low level computation on sensor data)
Transducer (conversion to signal) and Data Acquisition for measuring parameters
Source: Lebold and Thurston (2001)
Figure 4.
The OSA-CBM
framework
1230
IMDS
115,7
Downloaded by NATIONAL TECHNICAL UNIVERSITY OF ATHENS At 04:29 29 July 2015 (PT)
Next generation DSS for maintenance can use raw sensors events, domain knowledge
events, effect events, cause events and action events (Dunkel et al., 2011). Future
DSS should also include predictions and proactive actions based on these predictions
(Engel et al., 2012). Currently however, existing DSS in manufacturing support only
reactive event-driven applications as actions are taken after events have occurred.
A conceptual architecture for proactivity comprising predictive and proactive
agents for forecasting and automated decision making technologies, respectively,
has been proposed by Engel and Etzion (2011). Moreover, Artikis et al. (2014) presented
a conceptual methodology for scalable processing and event-driven decision making
which uses real-time optimization techniques in order to develop real-time proactive
planning tools applicable to traffic and credit card fraud management with various
levels of autonomy. However, there are opportunities for the development and
implementation of a holistic framework for providing novel information systems in
alignment with the e-maintenance needs of modern enterprises leveraging the potential
applications of proactivity and facilitating the way decisions are made.
3. Review of CBM methods and techniques
There are three main steps in CBM: diagnosis, prognosis and decision support
(Voisin et al., 2010; Peng et al., 2010). Diagnosis has to do with the actual monitoring of
a system and detection of failures, while prognosis has to do with prediction of the RUL
of the system based upon its actual health state (Venkatasubramanian, 2005). Although
diagnosis, which is usually mentioned as the step before prognosis, is not always
prerequisite, it can efficiently complement the proactive maintenance DSS in cases
when an undesired effect which has not been predicted occurs ( Jardine et al., 2006;
Peng et al., 2010).
Methods used for CBM can be classified in four categories (Venkatasubramanian,
2005; Goh et al., 2006): model-based; knowledge-based; data-driven; and combination of
them. In the current research work, data-driven methods are examined. However, they
are usually accompanied with some degree of knowledge depending on the availability
of data and the required output.
There are research works using methods for providing a diagnostic output,
a prognostic one or decision support, in other words they may stop to a different step
of CBM, based on Figure 3. For example, research works dealing with prognosis cover
the three first CBM steps of Figure 3, while those dealing with decision support cover
the decision support step as well. In other words, in the latter case, prognostic methods
provide a prediction based on which recommendations are generated.
Diagnosis and reactive recommendations (CAPA) is a well explored area (Ding et al.,
2002; Nandi et al., 2005; Jung et al., 2006; Qian et al., 2008; Bennouna and Roux, 2013;
Ruiz-Mezcua et al., 2011; Prakash and Ceglarek, 2013; Pal and Ceglarek, 2013;
Pal et al., 2014). For this reason, the focus of the current review is prognosis and
prognostic-based decision making.
The papers examined were identified by searching Google scholar with the
keywords “CBM,”“Condition Based Maintenance,”“recommendations,”“decision
support,”“decision making,”’manufacturing,”“maintenance,”“e-maintenance,”
“real-time”and “proactivity”in various combinations among them. We focussed on
papers dealing with the decision step of CBM. However, we realized that most of them
either were narrowed in the prognostic step of CBM (without reaching the decision step
for the provision of recommendations) or proposed a combination of methods so that
they develop a prognostic model based on real-time data and then, based on this, they
1231
Proactive
decision
making
framework
Downloaded by NATIONAL TECHNICAL UNIVERSITY OF ATHENS At 04:29 29 July 2015 (PT)
provide recommendations for maintenance. The focus was on most recent papers,
after 2008, with exceptions in cases where an older paper satisfied the keywords and
proposed a novel and useful method which has not been extended until now.
3.1 Prognosis
Voisin et al. (2010) focussed on the components of the prognostic process by focussing
on the prognosis sub-steps and illustrating its interactions with the other components
of the CBM strategy’s steps, as shown in Figure 5.
A significant body of research regarding the development of prognostic models has
been conducted. Several methods and techniques have been used in order to estimate
the RUL/ remaining life distribution (RLD) and/or the probability distribution about
the occurrence of a breakdown or other undesired events.
Banjevic and Jardine (2006) presented the failure process as a discrete Markov process
and Kolmogorov equation is used accompanied with product-integration method for the
calculation of RUL. Muller et al. (2008a) proposed a methodology which implements
the proactive logic in a prognosis model by combining probabilistic (dynamic Bayesian
networks –DBN) and event methods for degradation modeling and monitoring.
Salfner and Malek (2007) used HSMM in order to conduct online failure prediction
by using event-driven sources such as errors. The methodology was compared with
Dispersion Frame Technique, a reliability model and an event-based method in terms
of several performance measures such as precision, recall, F-measure, false-positive
rate and computing time. Gebraeel and Lawley (2008) developed a degradation model
based on condition monitoring with the use of NN. The model estimates and
continuously updates the RLD.
Gebraeel et al. (2009) presented a degradation modelling framework without
having historical data about degradation. So, they assume that failure time data follow
a Bernstein distribution to estimate the characteristics of the stochastic parameters
needed for the degradation modelling. Moreover, they assume that degradation
follows either a linear or an exponential distribution. The proposed methodology
estimates and continuously updates the RLD. Caesarendra et al. (2011) developed
a prognostic model based on statistical analysis to identify the actual degradation
of the component and to estimate the failure probability and its variance.
Prognostic
knowledge by
prognostic expert
Historical
data
To pilot
prognostics
Company Management (costs,
prognostic horizon, etc.)
To initialise state
and performances
To project
To compute RUL
RUL, confidence
level, future
performance
Diagnostic
Output
Source: Based on Voisin et al. (2010)
Figure 5.
Components of
prognostic process
1232
IMDS
115,7
Downloaded by NATIONAL TECHNICAL UNIVERSITY OF ATHENS At 04:29 29 July 2015 (PT)
Kim et al. (2012) presented a prognostic methodology for bearings of high pressure-
liquefied natural gas pumps which models the dynamic and stochastic degradation
process and estimates the RUL with the use of SVM. Tobon-Mejia et al. (2012a)
proposed a methodology for real-time data-driven prognosis with the use of a Mixture
of Gaussians- Hidden Markov Models (MoG-HMMs) and DBN, accompanied with other
techniques such as Baum-Welch algorithm, Viterbi algorithm, data clustering and
curve fitting, in order to calculate the RUL of the degrading machine tool and the
relevant confidence level.
Tobon-Mejia et al. (2012b) proposed a prognostic methodology for the estimation of
RUL and the relevant confidence level by using Wavelet Packet Decomposition
technique and the MoG-HMM as well as Baum-Welch algorithm and Viterbi algorithm.
Ferreiro et al. (2012) presented a framework for prognosis based on BNs embedded
to the IVHM concept. The output of the prognostic model is an estimation of RUL of the
component of the aircraft as well as its confidence values. The authors argue that this
prognostic information can contribute to the reduction of the costs caused by
cancellations or delays due to failures.
Bangalore and Tjernberg (2013) proposed a prognostic model based on ANN which
are updated continuously with the aid of an automated self-evolving approach and the
training data set is optimized. The model utilizes data taken from a Supervisory Control
and Data Acquisition (SCADA) system which is used for monitoring parts of
equipment.
Table I shows the prognostic methods that each paper uses accompanied with their
inputs and outputs.
Although a variety of methods have been used in order to provide useful
diagnostic and prognostic capabilities by utilizing real-time data from the
manufacturing domain such as the ones examined here, most have a low level
of autonomy because they narrow their decision support for human operators and
do not support partial or full autonomous decision making (Peng et al., 2010; Voisin
et al., 2010; Artikis et al., 2014).
3.2 Decision support
Several research works have examined and developed autonomous decision making
methods based on both historical and real-time data as well as expert knowledge with
the aim to address different maintenance challenges for components subjected to
condition monitoring.
Maintenance decision support is related to reliability, safety and environmental
issues as well as costs because of downtime of the equipment in case of a breakdown
or malfunctions of machines so it is a crucial operation function of the enterprise
(Peng et al., 2010). First, diagnostic and/or prognostic methods are applied and then,
the system recommends appropriate actions either for immediate implementation
due to an actual failure (reactive) or for future implementation in order to
avoid an undesired event (proactive actions). However, the latter is the least
explored area.
Several research works dealing with maintenance decision support are based on
predictions. Predictions are not considered as given in these works, e.g., as prognostic
functions, probabilistic estimates or expert knowledge. Hence, research works dealing
with decision support usually develop a prognostic model analyzing and processing the
historical and real-time data available and, based on these, they develop decision
methods in order to provide prognostic-based recommendations.
1233
Proactive
decision
making
framework
Downloaded by NATIONAL TECHNICAL UNIVERSITY OF ATHENS At 04:29 29 July 2015 (PT)
References Input Prognostic methods Output
Muller et al. (2008a) Real-time data: degradation
Historical data: historical failure/
degradation data
Knowledge: process knowledge
Dynamic Bayesian networks (probabilistic model)
Event model
Probability distribution over all variables
Consistency of prognosis
Banjevic and
Jardine (2006)
Real-time data: degradation
(e.g. vibration or oil)
Historical data: historical failure/
degradation data
Knowledge: wear stages
Degradation Modelling
Markov Chain
Reliability functions
Calculation of RUL as a function of the
current conditions
Salfner and Malek
(2007)
Real-time data: real time monitoring
of error logs
Historical data: recorded error logs
Hidden Semi-Markov Models (HSMMs) (that use
event-driven sources such as errors)
Forecasting of the occurrence of failures
Gebraeel and
Lawley (2008)
Real-time data: degradation signals
(e.g. vibration signals)
Historical data: historical
degradation data (e.g. vibration)
Neural Network
Degradation Modelling
Computing residual life distribution
Updating residual life distribution
Gebraeel et al.
(2009)
Real-time data: degradation signals
(e.g. vibration signals)
Historical data: historical failure
time data
Knowledge: wear stages
Degradation modelling
Bayesian networks
Residual life prediction
Caesarendra et al.
(2011)
Real-time data: vibration condition
monitoring data
Historical data: bearing failure data
Relevance vector machine
Logistic regression
Autoregressive moving average
Dempster-Shafer regression
Statistical process control
Prediction of the failure probability of
individual units of bearing samples
Analysis of variance of the failure probability
Kim et al. (2012) Real-time data: vibration signals
Historical data: historical
degradation (vibration) data
Knowledge: wear stages
Support vector machine (SVM) classifier Optimal prediction of RUL
(continued )
Table I.
Reviewed research
works on prognosis
1234
IMDS
115,7
Downloaded by NATIONAL TECHNICAL UNIVERSITY OF ATHENS At 04:29 29 July 2015 (PT)
References Input Prognostic methods Output
Tobon-Mejia et al.
(2012a)
Real-time data: degradation data
Historical data: degradation data
Knowledge: wear stages
Mixture of Gaussian-Hidden Markov model
Dynamic Bayesian network
Baum-Welch algorithm
Viterbi algorithm
RUL and confidence value
Tobon-Mejia et al.
(2012b)
Real-time data: degradation signals
(e.g. vibration signals)
Historical data: historical
degradation data (e.g. vibration)
Knowledge: wear stages
Mixture of Gaussian-Hidden Markov model
Dynamic Bayesian network
Wavelet packet decomposition
Baum-Welch algorithm
Viterbi algorithm
RUL and confidence value
Ferreiro et al.
(2012)
Real-time data: degradation signals
Historical data: historical
degradation data
Knowledge: domain knowledge
about causes and effects
Bayesian networks Time to failure
RUL and confidence value
Bangalore and
Tjernberg (2013)
Real-time data: signals from SCADA
system
Knowledge/historical data: normal
behaviour of components
Artificial neural network Fault prognosis
Table I.
1235
Proactive
decision
making
framework
Downloaded by NATIONAL TECHNICAL UNIVERSITY OF ATHENS At 04:29 29 July 2015 (PT)
Kaiser and Gebraeel (2009) proposed a method for predictive maintenance management
by utilizing real-time degradation data. They developed a degradation model in order to
estimate and update in real time the RLD of some part of equipment and the most
suitable maintenance policy is recommended based on the frequency of failures and the
maintenance costs.
Besnard and Bertling (2010) presented a method for applying CBM to wind turbine
blades. Provided that degradation of a part of equipment can be classified into one
category in terms of severity of damage, the method optimizes decisions about different
maintenance strategies. The strategies examined in this paper are visual inspection,
inspection with a condition-monitoring technique and online condition monitoring.
Besnard et al. (2011) presented a model for the optimization of maintenance planning
in offshore wind farms. The model uses stochastic optimization in order to perform the
optimum maintenance schedule at the lowest cost based upon the wind and production
forecasting.
Castro et al. (2012) proposed a cost modelling approach for predictive maintenance
policy based on the RUL estimated after each inspection. The method recommends the
optimal time of applying maintenance to avoid a future breakdown. The failures that
are assumed to happen can depend on degradation or on immediate shocks of the
equipment.
Wu et al. (2007) developed a DSS in order to minimize the expected cost by taking
into account the RUL. The authors propose an ANN to predict and update in real-time
the RUL of the equipment and cost modelling techniques accompanied with probability
theory for calculating the replacement time which minimizes cost at each unit
operational time.
Ivy and Nembhard (2005) proposed a method for recommending the optimal
maintenance policy by using statistical quality control (SQC) and partially observable
Markov decision process (POMDP). SQC techniques were used in order to define the
observations distributions and the structure of POMDP. Results of POMDP were
evaluated in terms of robustness and accuracy.
Aissani et al. (2009) used reinforcement learning and Markov decision process
(MDP) in order to automate maintenance tasks scheduling in petroleum industry and to
update them in real time. The model developed has multiple agents which function in
the logic of continuous improvement by learning the best behaviours of their roles and
improving the solution about the corrective and predictive tasks provided.
Elwany and Gebraeel (2008) proposed a decision model for component replacement
and spare parts inventory based upon the RLD instead of failure time distribution. RLD
is calculated and is continuously updated and is then used as input to the decision
model which calculates the optimal replacement time as well as the optimal inventory
ordering time.
Bouvard et al. (2011) proposed a method for the optimization of maintenance
planning for systems with multiple components such as commercial heavy vehicles.
First, maintenance actions are grouped according to the component to which they are
linked. Then, degradation models are developed in order to monitor each component,
recommend the optimal maintenance planning and update it dynamically when
necessary.
Huynh et al. (2012) proposed a method for assessing, comparing and selecting the
most appropriate and cost-effective maintenance policy for a single-unit degrading
system under condition monitoring. Muller et al. (2007) proposed a methodology which
implements the proactive logic in a prognosis model for supporting the maintenance
1236
IMDS
115,7
Downloaded by NATIONAL TECHNICAL UNIVERSITY OF ATHENS At 04:29 29 July 2015 (PT)
strategy. This evolution of prognostic models combines probabilistic and event
methods in order to evaluate different maintenance plans in terms of effectiveness and
cost and to calculate the optimal maintenance policy.
Engel et al. (2012) provided a methodology for proactive event-driven computing with
potential application to CBM, however, it is conceptually described and sets the guidelines
for future development. Suitable methods for proactive applications could be BN and MDP.
Table II summarizes the prognostic-based decision support methods reviewed,
as well as their inputs and outputs. The methods have been separated in two groups;
one group supports the prognostic and the second the decision step. The prognostic
step of CBM is the one that contributes to the input of the decision support step.
Despite the advances in CBM, limitations and open issues still exist, such as
( Jardine et al., 2006; Iung et al., 2009; Peng et al., 2010):
•Prognostic models used for CBM are not always continuously updated by
real-time data through sensors, but they receive batches of data. This fact affects
negatively the responsiveness of the system to provide prognostic information
and recommendations for maintenance. The reason for this is that high
computation speed is required, so an appropriate system needs to be developed.
•Although there are several theoretical research works, there is a limited number
of practical applications.
•Recommendations of maintenance actions and maintenance policy for CBM are
not usually embedded in integrated maintenance platforms.
•Despite the plethora of existing works for both prognosis and diagnosis in
maintenance, most of them do not examine automation of decisions by providing
recommendations for maintenance actions.
•E-maintenance could be a significant contributor to the conversion
of maintenance from reactive to proactive.
•Collecting all necessary data (both historical and real-time) in order to develop
a method which provides reliable results is a major challenge.
4. Proactive decision making framework for CBM
In this section we present a conceptual framework for supporting CBM decisions
within an e-maintenance/real-time data infrastructure. As shown in Figure 6, our
framework represents the sequence of steps, which need to be followed in order to
support decision making in e-maintenance. The framework is based on the OSA-CBM
framework (Lebold and Thurston, 2001) and the prognostic process outlined by
Voisin et al. (2010), as outlined in Section 2. The conceptual framework of Figure 6
extends the aforementioned works in two ways: first, the CBM constituents are
identified and structured in two categories and second, the decision support constituent
is further analyzed according to the types of decisions that can be provided. Specifically,
the two categories are the information space and the decision space. The former
consists of diagnosis and prognosis and provides information about the current and
the future health state of the equipment, respectively, while the latter consists of
maintenance actions, both reactive and proactive ones. An integrated view of the
information and decision spaces is a prerequisite for providing timely and reliable
recommendations because the input of the decision space relies on the predictions made
within the information space.
1237
Proactive
decision
making
framework
Downloaded by NATIONAL TECHNICAL UNIVERSITY OF ATHENS At 04:29 29 July 2015 (PT)
References Input Prognostic methods Output Input Decision methods Output
Kaiser and
Gebraeel (2009)
Real-time data
Vibration
Historical data
Failure times
Degradation
Downtime
Continuous-time
continuous-state
stochastic model
Degradation
modelling
Estimation of RLD RLD
Knowledge
Costs (planned
replacements and total
maintenance costs)
Process knowledge
Cost modelling
Rules
Compute/update/
evaluate maintenance
schedule
Besnard and
Bertling (2010)
Real-time data
Degradation
Knowledge
Lifetime
Failures
States
Degradation
modelling
Markov chain
Sensitivity analysis
RUL
Failure rate-crack
initiation rate
Mean crack time to
failure
RUL
Failure rate
Average production
Maintenance and
production costs
Continuous time
Markov chain
optimization Rules
Optimal maintenance
strategy
Besnard et al.
(2011)
–––Knowledge
Wind forecasting
Failure rate
List of actions
Repair time
Maintenance and
logistics costs
Stochastic
optimization
Rules
Cost for production
losses and
transportation that
could be saved
Castro et al.
(2012)
Real-time data
Degradation
Historical data
Degradation
Knowledge
Threshold limit
Degradation
modelling
Statistical analysis
Mean residual life
Probability of
preventive and
corrective
replacement
Replacement time
Mean residual life
Probability of preventive
and corrective
replacement
Expected downtime
Replacement time
Knowledge
Costs of preventive and
corrective replacement
Cost of inspection
Optimization
(considering both
degradation and
traumatic shocks)
Minimized long-run
expected
maintenance cost
Optimum policy
(continued )
Table II.
Reviewed research
works on
prognostic-based
decision support
1238
IMDS
115,7
Downloaded by NATIONAL TECHNICAL UNIVERSITY OF ATHENS At 04:29 29 July 2015 (PT)
References Input Prognostic methods Output Input Decision methods Output
Wu et al. (2007) Real-time data
Vibration
Historical data
Vibration
Knowledge
Failure threshold
Artificial neural
network
Non-linear
programming
(Levenberg-
Marquardt
algorithm)
Moving average
Residual life
percentile prediction
Marginal residual
life distribution
Predicted residual life
percentile
Marginal residual life
distribution
Knowledge
Operating time
Corrective and predictive
maintenance costs
Non-linear
programming
Cost matrix/
expected cost
optimization
Predicted failure time
Minimized cost
Optimal replacement
time
Ivy and
Nembhard
(2005)
Real-time data
Degradation
Knowledge
States
Probabilities
Threshold
Statistical quality
control (SQC)
Transition matrix
Estimation of
distribution
parameters
Transition matrix
Estimation of the
distribution parameters
Total expected cost
Maintenance costs
Partially observable
Markov decision
process (POMDP)
Minimum total
expected cost
Maintenance actions
Aissani et al.
(2009)
Real-time data
Failures
Historical data
Failures
Knowledge
States
Operational times
Actions
Reinforcement
learning (SARSA
algorithm): solve
selection and reward
function
Solution of reward
function
Solution of selection
function
Probabilities of
occurrence of events
Solution of reward
function
Solution of selection
function
Probabilities of occurrence
of events
Markov decision
process
Generate online
scheduling solutions
for predictive and
corrective
maintenance tasks
on-line
Elwany and
Gebraeel (2008)
Real-time data
Vibration
Historical data
Vibration
Degradation
modelling
RLD RLD
Knowledge
Planned and failure
replacement cost
Holding and shortage
costs
Lead times
Optimization
(replacement model)
Optimal replacement
and inventory
ordering times
(continued )
Table II.
1239
Proactive
decision
making
framework
Downloaded by NATIONAL TECHNICAL UNIVERSITY OF ATHENS At 04:29 29 July 2015 (PT)
References Input Prognostic methods Output Input Decision methods Output
Bouvard et al.
(2011)
Real-time data
Degradation
Knowledge
Deterioration
parameters
Degradation
modelling
Statistics
Failure probability
function
Degradation path
Time-to-failure
Failure probability
function
Estimated degradation
path
Time-to-failure
Knowledge
Maintenance costs
Maintenance
optimization
Optimal grouping
structure
Optimal maintenance
dates and costs
Huynh et al.
(2012)
Real-time data
Periodic inspection
Knowledge
Degradation
process
Threshold
Degradation
modelling
Statistics
particle filtering
Condition reliability
Measurement of
uncertainty
Probability density
function
Condition reliability
Measurement of
uncertainty
Probability density
function
Knowledge
Cost function
Optimization-
dynamic
replacement model
Rules
Predictive
replacement time
estimation
Optimized cost
Muller et al.
(2007)
Real-time data
Failures
Historical data
Failures
Knowledge
Process
Dynamic Bayesian
networks
Event model
Probability
distribution over all
variables
Consistency of
prognosis
Probability distribution
over all variables
Consistency of prognosis
Knowledge
List of actions
Costs
DBN
Utility function
Multi-criteria
analysis
Markov chain
Assessment of
maintenance
alternatives
Optimal maintenance
policy
Engel et al.
(2012)
Real-time data
Failures
Historical data
Failures
Bayesian networks Probability
distribution of time
to breakdown
Probability distribution
Time to breakdown
Knowledge
States
Actions
Cost function
Markov decision
process
Optimal action
Optimal time of
action
Table II.
1240
IMDS
115,7
Downloaded by NATIONAL TECHNICAL UNIVERSITY OF ATHENS At 04:29 29 July 2015 (PT)
In the general case, all steps shown in Figure 6 should be followed; however
there may be limitations regarding the availability of data, which may hinder
some steps. For example, if there is no list of actions and their mapping to
types of failures or defects, or there is lack of other information regarding maintenance
strategy or schedule, the framework can only provide prognostic information
and cannot provide automated support for making decisions about the strategy,
schedule or action.
Two main steps are included within the information space: diagnosis; and
prognosis. Diagnosis involves a sequence of three sub-steps which represent the
required analysis of the raw sensor data that are gathered until the detection of the
current equipment health state. These sub-steps are:
(1) Signal processing, which provides some initial and primitive computation of
sensor data.
(2) Condition monitoring, which compares the results of signal processing data
analysis with predefined features.
(3) Health assessment, which combines condition monitoring data and historical
data in order to provide information about the actual health state of the system
examined (fault detection).
Information Space Decision Space
Reactive Recommendations
Recommendations about CAPA
Proactive Recommendations
Recommendations about optimal
time for a predefined action
Recommendations about optimal
action and time
Recommendations about
maintenance strategy or schedule
Recommendations about
maintenance strategy or schedule
Diagnosis
Signal Processing
Condition Monitoring
Health assessment
Prognosis
Figure 6.
Framework for
decision making in
maintenance
1241
Proactive
decision
making
framework
Downloaded by NATIONAL TECHNICAL UNIVERSITY OF ATHENS At 04:29 29 July 2015 (PT)
In case of recommendations for reactive actions, the next step would be
recommendations about CAPA or maintenance strategy or schedule. In other words,
based on the information about the current health state provided by the information
space, the decision space focusses on generating the recommendations for mitigating
actions. Support for reactive actions should be embedded even within a proactive
DSS, because there is always the possibility that an undesired event, which has
not been predicted, occurs. In this case, diagnostic information should be provided and
maintenance actions for immediate implementation should be recommended.
Having examined the health state of the system, a prognostic model should be
developed in order to support proactive recommendations. The prognostic model
combines the diagnostic information and historical degradation data and patterns
leading to a failure as well as domain knowledge and represents progression of system
health. Domain knowledge can be modelling wear stages, degradation threshold
limit, failure mode effects and criticality analysis, root cause analysis, fault tree
analysis, etc. For example, relationships between effects such as failure and
malfunction, and sensor parameters such as temperature and vibration should
be identified. In this way, predictions can be made for when a failure will occur,
calculating probability distributions of the occurrence of undesired events (e.g. failure if
no action is implemented, failure even though an action has been implemented, etc.)
and/or providing early warnings.
The output of the prognostic model feeds into the proactive decision making process,
which is represented by the Proactive Recommendations constituent of our conceptual
framework’s Decision Space. Proactive decision making utilizes domain knowledge
(e.g. list of actions, time intervals for each possible action, cost function, etc.) and
recommends the optimal maintenance strategy, the optimal schedule, the optimal time of
applying a predefined action or the optimal action and its time of applying. The output of
proactive decision making depends on the user requirements as well as the available data
and knowledge and it ranges from generic (e.g. maintenance strategy or schedule) to
specific recommendations (e.g. maintenance action and time of applying it).
Each step of the decision making framework shown in Figure 6 requires specific input
and provides specific output which feeds into the next step. Input depends on the
availability of the appropriate data and information and output can vary based upon the
user’s requirements, the method used and the taken input. Table III summarizes the input
and output of each step of the decision making framework based on our literature review.
Our conceptual framework extends the existing works in CBM in two ways. First,
it identifies the CBM constituents and organizes them in the Information Space and the
Decision Space. The Information Space includes the diagnosis phase dealing with
the provision of information about the current state of equipment, and the prognosis
phase dealing with the provision of information about the future health state
of equipment. This information supports informed decision-making about maintenance
in the sense that it reveals issues that are not visible even by an experienced engineer.
Second, while the Information Space has been extensively researched, our literature
review revealed that the Decision Space has not been investigated thoroughly.
Therefore, we further analysed the decision support constituent according to the types
of recommendations that can be provided. The latter have been separated in two types:
reactive and proactive recommendations. Each one of these types has been further
analysed according to the provided output as shown in Table III. Reactive
recommendations deal with actions that are implemented after the occurrence of an
undesired event (e.g. breakdown). We found out that these actions can involve either
1242
IMDS
115,7
Downloaded by NATIONAL TECHNICAL UNIVERSITY OF ATHENS At 04:29 29 July 2015 (PT)
the implementation of a CAPA (Qian et al., 2008; Peng et al., 2010; Ruiz-Mezcua et al., 2011;
Prakash and Ceglarek, 2013; Pal and Ceglarek, 2013; Pal et al., 2014) or a change in the
maintenance strategy or schedule. On the other hand, proactive recommendations are
based on predictions about an undesired event (e.g. a future breakdown). We found
out that proactive recommendations can be separated in three categories, according to
their output which can be: a change in maintenance strategy or maintenance schedule
(Muller et al., 2007; Kaiser and Gebraeel, 2009; Aissani et al., 2009; Besnard and Bertling,
2010; Besnard et al., 2011), the optimal time of applying a predefined action
(e.g. replacement) (Wu et al., 2007; Elwany and Gebraeel, 2008; Bouvard et al., 2011;
Castro et al., 2012; Huynh et al., 2012) or pairs of optimal actions and optimal time for
their implementation (Engel et al., 2012).
5. Practical demonstration of the proposed framework
In this section we present a practical application of the proposed proactive decision
making framework for CBM in the oil and gas industry. We describe the practical
role and use of the proposed framework focussing on how it can support proactive
decision-making ahead of time on the basis of real-time observations and predictions
about future undesired events, through a real maintenance scenario.
Input Output
Diagnosis Sensor data (about the measured parameter
used as indicator of degradation)
Historical data (about the measured parameter
used as indicator of degradation till failure)
Current health state
Reactive
actions
Current health state
List of actions (that are integrated with the
current undesired event) or predefined action or
alternative strategies
Notification
Recommendations about
CAPA
Maintenance strategy or schedule
Prognosis Current health state
Sensor data (about the measured parameter
used as indicator of degradation)
Historical data (about the measured parameter
used as indicator of degradation till failure)
Threshold limit (where a failure/ malfunction
occurs)
Wear stages (from domain knowledge)
Other domain knowledge
Early warnings
RUL and confidence level
Probability distributions of the
occurrence of undesired (e.g. failure,
malfunction, etc.)
Proactive
actions
Early warnings
RUL and confidence level
Probability distributions of the occurrence of
undesired events (e.g. failure, malfunction, etc.)
List of actions (that mitigate or eliminate the
future undesired event) or predefined action or
alternative strategies
Cost functions (for the forecasted event
and for the possible actions)
Time intervals (delays) for each
possible action
Early notification
Recommendations about
Maintenance strategy or schedule
Optimal time for a predefined action
Optimal action and time
Table III.
Input and output in
each step of the
decision making
framework
1243
Proactive
decision
making
framework
Downloaded by NATIONAL TECHNICAL UNIVERSITY OF ATHENS At 04:29 29 July 2015 (PT)
CBM in the oil and gas industry employs various monitoring means to detect
deterioration and failure in some critical drilling equipment. In our example, we focus
on the gearbox drilling equipment and consider as indicators the rotation speed of the
drilling machine’s main shaft in Rounds Per Minute (RPM), along with the lube
oil temperature of the drilling machine’s gearbox. Temperature sensors gathering
real-time data in a high frequency (every 20 ms), along with historical data of oil
temperature, RPM events and gearbox equipment failure, are used for assessing the
health state of the gearbox in real-time (see Figure 6). The high frequency of the
real-time data requires a Big Data infrastructure and an appropriate architecture
(e.g. EDA). Diagnosis step identifies that a drilling gearbox equipment failure starts to
occur and, therefore, informs about its actual health state (e.g. anomalies detection).
The prognosis step involves the usage of statistical/machine learning methods to
build a prognosis model of the equipment health offline, as well as the online prediction
of gearbox’s RUL along with the probability distribution function of the gearbox
breakdown by using methods such as Bayesian networks, neural networks, etc.
The prognostic output (e.g. a specific exponential probability distribution) is needed as
input for the proactive decision making process accompanied with a list of alternative
maintenance actions (lube oil change, system restart, lower pressure, full maintenance),
the cost of each action as a function of time, the cost of gearbox breakdown, the time of
next planned maintenance and the RUL after the implementation of each action.
A recommendation about the optimal action and the optimal time for its implementation is
provided by employing decision making methods such as MDP. In this way, the oil and
gas company is able to know which action to do and when. To eliminate the risk that an
undesired event, which has not been predicted, occurs, the company has also defined
based on domain knowledge, a list of reactive actions in the form of IF-THEN rules.
In this way, the company is guided about how to utilize its sensors and historical
data as well as its domain knowledge in order to improve its maintenance management
business process and turn from reactive to proactive for the improvement of its
efficiency.
6. Discussion
Although decision making methods based on predictions about the health state of the
equipment has been proposed in the last years, not all of the capabilities of proactive
computing have been exploited. The proposed framework allows embedding in the
CBM concept e-maintenance capabilities in conjunction with proactive decision making
in order to allow the provision of more detailed recommendations. In this way,
the decision maker is based less on his/her personal judgement and is able to take
proactively informed decisions. At the same time, the time window for taking decisions
about how to resolve a problem is increased, as the decision epoch starts upon
the prediction of the problem instead of upon its occurrence. So, on the one hand the
probability of human error is decreased while, on the other hand, there is more time for
planning and undertaking other manufacturing operations that are closely related
to maintenance (e.g. logistics issues, such as the ordering of spare parts). Furthermore,
based on the proposed framework, the optimal time for maintenance can be
accompanied with the appropriate action according to the predictions, the user
requirements and the company policies. Therefore, the higher the level of proactivity is
achieved, the more efficient the maintenance becomes, because undesired events
are eliminated or mitigated and all the business functions that are related to
maintenance, such as production, ordering of spare parts, etc., operate seamlessly.
1244
IMDS
115,7
Downloaded by NATIONAL TECHNICAL UNIVERSITY OF ATHENS At 04:29 29 July 2015 (PT)
The systematic formulation of CBM strategy in the proposed framework enables its
embodiment to an information system for the full exploitation of proactive decision
making in the context of CBM. In this sense, the proposed framework is not only
a systematic representation of a maintenance management business process but also
the basis for the development of a DSS for CBM that gathers and analyses real-time
sensor data, provides diagnostic and prognostic information and automates proactive
decisions by providing recommendations about maintenance in a proactive manner.
Real-time data are gathered in a high frequency, so several challenges regarding
Big Data need to be addressed by applying innovative technologies. EDA can
significantly enable the processing of Big Data, so that predictions are provided when
an anomaly detection event is received and recommendations are generated when
an undesired event (e.g. breakdown) is predicted. As the decision space of the proposed
framework has been structured according to the types of recommendations that can be
provided, a DSS that is built upon it will be able to support proactive decision making
for CBM in various application domains and for a wide range of functional and non-
functional application requirements. Provision of reactive recommendations should
be also made available because the probability that the DSS fails to predict an
undesired event cannot be completely eliminated; in that case immediate actions to
handle the undesired event should be recommended.
The focus of the literature search and the whole analysis was on proactive decision
making rather than reactive and consequently, on data-driven prognosis and proactive
recommendations. For this reason, the literature for methods dealing with the three
sub-steps of Diagnosis has not been examined in detail.
7. Conclusions and future work
Our literature review of CBM methods indicates that there are combinations of machine
learning and decision making methods used in order to provide recommendations
based on predictions. These predictions are derived from the analysis and processing
of real-time sensor data. For each paper examined, the methods as well as their inputs
and outputs were identified. Based on previous research works regarding modeling
CBM concept and on the literature review of methods and techniques that are used for
prognosis and decision making, a framework for maintenance decision making is
proposed. This framework enriches others existing in literature by structuring the
information and the decision space and by focussing on the latter. Moreover,
the proposed framework embeds the concepts of proactivity and e-maintenance to CBM
in order to enable the provision of timely and reliable recommendations. Finally,
reactive actions for immediate implementation based on diagnostic information are
recommended when an undesired event that has not been predicted occurs.
Although there are many research works dealing with predictions, e.g. about RUL,
only few propose methods to utilize this real-time prediction accompanied with expert
knowledge to provide maintenance recommendations. More combinations of methods
can be developed by using machine learning methods that have been used in literature
for prediction of RUL with various decision methods so that they are extended in
the Decision Space of the framework proposed. Furthermore, decision methods in
existing literature do not exploit all the possibilities for proactive decision making,
for example by providing the most appropriate maintenance action and the optimal
time for applying it, while, they are not usually considered as part of a wider
framework for decision making in CBM.
1245
Proactive
decision
making
framework
Downloaded by NATIONAL TECHNICAL UNIVERSITY OF ATHENS At 04:29 29 July 2015 (PT)
Based on the proposed framework, we plan to develop a DSS which will provide
proactive maintenance recommendations according to user requirements. Then, we will
examine the possibility of using context-awareness so that data are enriched and
recommendations take into account the various conditions that may affect them.
References
Aissani, N., Beldjilali, B. and Trentesaux, D. (2009), “Dynamic scheduling of maintenance tasks in
the petroleum industry: a reinforcement approach”,Engineering Applications of Artificial
Intelligence, Vol. 22 No. 7, pp. 1089-1103.
Artikis, A., Baber, C., Bizarro, P., Canudas-de-Wit, C., Etzion, O., Fournier, F., Goulart P., Howes, A.,
Lygeros, J., Paliouras, G., Sharfman, I. and Schuster, A. (2014), “Scalable proactive
event-driven decision-making”,Technology and Society Magazine, IEEE, Vol. 33 No. 3,
pp. 35-41.
Bangalore, P. and Tjernberg, L.B. (2013), “An approach for self-evolving neural network based
algorithm for fault prognosis in wind turbine”,2013 IEEE Grenoble PowerTech
(POWERTECH), IEEE, Vol. 1 No. 1, pp. 1-6.
Banjevic, D. and Jardine, A.K.S. (2006), “Calculation of reliability function and remaining
useful life for a Markov failure time process”,IMA Journal of Management Mathematics,
Vol. 17 No. 2, pp. 115-130.
Benedettini, O., Baines, T.S., Lightfoot, H.W. and Greenough, R.M. (2009), “State-of-the-art in
integrated vehicle health management”,Proceedings of the Institution of Mechanical
Engineers, Part G: Journal of Aerospace Engineering, Vol. 223 No. 2, pp. 157-170.
Bennouna, O. and Roux, J.P. (2013), “Real time diagnosis & fault detection for the reliability
improvement of the embedded systems”,Journal of Signal Processing Systems, Vol. 73
No. 2, pp. 153-160.
Besnard, F. and Bertling, L. (2010), “An approach for condition-based maintenance optimization
applied to wind turbine blades”,Sustainable Energy, IEEE Transactions On, Vol. 1 No. 2,
pp. 77-83.
Besnard, F., Patriksson, M., Stromberg, A., Wojciechowski, A., Fischer, K. and Bertling, L. (2011),
“A stochastic model for opportunistic maintenance planning of offshore wind farms”,2011
IEEE Trondheim PowerTech, IEEE, pp. 1-8.
Bouvard, K., Artus, S., Berenguer, C. and Cocquempot, V. (2011), “Condition-based dynamic
maintenance operations planning & grouping, application to commercial heavy vehicles”,
Reliability Engineering & System Safety, Vol. 96 No. 6, pp. 601-610.
Caesarendra, W., Widodo, A., Thom, P.H., Yang, B.S. and Setiawan, J.D. (2011), “Combined
probability approach and indirect data-driven method for bearing degradation
prognostics”,IEEE Transactions On Reliability, Vol. 60 No. 1, pp. 14-20.
Castro, I.T., Huynh, K.T., Barros, A. and Berenguer, C. (2012), “A predictive maintenance strategy
based on mean residual life for systems subject to competing failures due to degradation
and shocks”,Proceedings of the 11th International Probabilistic Safety Assessment and
Management Conference & the Annual European Safety and Reliability Conference-PSAM
11/ESREL 2012, pp. 375-384.
Ding, Y., Ceglarek, D. and Shi, J. (2002), “Fault diagnosis of multistage manufacturing processes
by using state space approach”,Journal of Manufacturing Science and Engineering,
Vol. 124 No. 2, pp. 313-322.
Dunkel, J., Fernández, A., Ortiz, R. and Ossowski, S. (2011), “Event-driven architecture for
decision support in traffic management systems”,Expert Systems with Applications, Vol. 38
No. 6, pp. 6530-6539.
1246
IMDS
115,7
Downloaded by NATIONAL TECHNICAL UNIVERSITY OF ATHENS At 04:29 29 July 2015 (PT)
Dunsdon, J. and Harrington, M. (2008), “The application of open system architecture for condition
based maintenance to complete IVHM”,2008 IEEE Aerospace Conference, IEEE Vol. 1
No. 1, pp. 1-9.
Elwany, A.H. and Gebraeel, N.Z. (2008), “Sensor-driven prognostic models for equipment
replacement and spare parts inventory”,IIE Transactions, Vol. 40 No. 7, pp. 629-639.
Engel, Y. and Etzion, O. (2011), “Towards proactive event-driven computing”,Proceedings of the
5th ACM International Conference on Distributed Event-Based System, ACM, pp. 125-136.
Engel, Y., Etzion, O. and Feldman, Z. (2012), “A basic model for proactive event-driven
computing”,Proceedings of the 6th ACM International Conference on Distributed Event-
Based Systems, ACM, pp. 107-118.
Ferreiro, S., Arnaiz, A., Sierra, B. and Irigoien, I. (2012), “Application of bayesian networks in
prognostics for a new integrated vehicle health management concept”,Expert Systems with
Applications, Vol. 39 No. 7, pp. 6402-6418.
Garg, A. and Deshmukh, S.G. (2006), “Maintenance management: literature review and
directions”,Journal of Quality in Maintenance Engineering, Vol. 12 No. 3, pp. 205-238.
Gebraeel, N., Elwany, A. and Pan, J. (2009), “Residual life predictions in the absence of prior
degradation knowledge, IEEE Transactions on Reliability, Vol. 58 No. 1, pp. 106-117.
Gebraeel, N.Z. and Lawley, M.A. (2008), “A neural network degradation model for computing and
updating residual life distributions”,IEEE Transactions on Automation Science and
Engineering, Vol. 5 No. 1, pp. 154-163.
Goh, K.M., Tjahjono, B., Baines, T. and Subramaniam, S. (2006), “A review of research in
manufacturing prognostics”,2006 IEEE International Conference on, Industrial
Informatics, IEEE, pp. 417-422.
Huynh, K.T., Barros, A. and Berenguer, C. (2012), “Maintenance decision-making for systems
operating under indirect condition monitoring: value of online information and impact of
measurement uncertainty”,IEEE Transactions on Reliability, Vol. 61 No. 2, pp. 410-425.
Irigaray, A.A., Gilabert, E., Jantunen, E. and Adgar, A. (2009), “Ubiquitous computing for
dynamic condition-based maintenance”,Journal of Quality in Maintenance Engineering,
Vol. 15 No. 2, pp. 151-166.
Iung, B., Levrat, E., Marquez, A.C. and Erbe, H. (2009), “Conceptual framework for e-maintenance:
illustration by e-maintenance technologies and platforms”,Annual Reviews in Control,
Vol. 33 No. 2, pp. 220-229.
Ivy, J.S. and Nembhard, H.B. (2005), “A modeling approach to maintenance decisions using
statistical quality control and optimization”,Quality and Reliability Engineering
International, Vol. 21 No. 4, pp. 355-366.
Jardine, A.K., Lin, D. and Banjevic, D. (2006), “A review on machinery diagnostics and
prognostics implementing condition-based maintenance”,Mechanical Systems and Signal
Processing, Vol. 20 No. 7, pp. 1483-1510.
Jung, J.H., Jong-Jae, L. and Kwon, B.H. (2006), “Online diagnosis of induction motors using
MCSA”,IEEE Transactions on Industrial Electronics, Vol. 53 No. 6, pp. 1842-1852.
Kaiser, K.A. and Gebraeel, N.Z. (2009), “Predictive maintenance management using sensor-based
degradation models”,IEEE Transactions on Systems, Man and Cybernetics, Part A:
Systems and Humans, Vol. 39 No. 4, pp. 840-849.
Kim, H.E., Tan, A.C., Mathew, J. and Choi, B.K. (2012), “Bearing fault prognosis based on health
state probability estimation”,Expert Systems with Applications, Vol. 39 No. 5, pp. 5200-5213.
Lebold, M. and Thurston, M. (2001), “Open standards for condition-based maintenance and
prognostic systems”,5th Annual Maintenance and Reliability Conference (MARCON 2001),
Gatlinburg, TN, May 25-27.
1247
Proactive
decision
making
framework
Downloaded by NATIONAL TECHNICAL UNIVERSITY OF ATHENS At 04:29 29 July 2015 (PT)
Lee, J., Abujamra, R., Jardine, A.K., Lin, D. and Banjevic, D. (2004), “An integrated platform for
diagnostics, prognostics and maintenance optimization”,Proceedings of the Intelligent
Maintenance Systems, pp. 15-27.
Lee, J., Ni, J., Djurdjanovic, D., Qiu, H. and Liao, H. (2006), “Intelligent prognostics tools and
e-maintenance”,Computers in Industry, Special Issue on e-Maintenance, Vol. 57 No. 6,
pp. 476-489.
Léger, J.B. and Morel, G. (2001), “Integration of maintenance in the enterprise: towards an
enterprise modeling-based framework compliant with proactive maintenance strategy”,
Production Planning and Control, Vol. 12 No. 2, pp. 176-187.
Levrat, E. and Iung, B. (2007), “TELMA: a full e-maintenance platform”,Proceedings of the Second
World Congress on Engineering Asset Management (WCEAM), Harrogate, June 11-14.
Macchi, M., Márquez, A.C., Holgado, M., Fumagalli, L. and Martínez, L.B. (2014), “Value-driven
engineering of e-maintenance platforms”,Journal of Manufacturing Technology
Management, Vol. 25 No. 4, pp. 568-598.
Magoutas, B., Stojanovic, N., Bousdekis, A., Apostolou, D., Mentzas, G. and Stojanovic, L. (2014),
“Anticipation-driven architecture for proactive enterprise decision making”,
CAiSE (Forum/Doctoral Consortium), pp. 121-128.
Muller, A., Crespo Marquez, A. and Iung, B. (2008b), “On the concept of e-maintenance: review
and current research”,Reliability Engineering & System Safety, Vol. 93 No. 8, pp. 1165-1187.
Muller, A., Suhner, M.C. and Iung, B. (2007), “Maintenance alternative integration to
prognosis process engineering”,Journal of Quality in Maintenance Engineering, Vol. 13
No. 2, pp. 198-211.
Muller, A., Suhner, M.C. and Iung, B. (2008a), “Formalisation of a new prognosis
model for supporting proactive maintenance implementation on industrial system”,
Reliability Engineering & System Safety, Vol. 93 No. 2, pp. 234-253.
Nandi, S., Toliyat, H.A. and Li, X. (2005), “Condition monitoring and fault diagnosis of electrical
motors-a review”,IEEE Transactions on Energy Conversion, Vol. 20 No. 4, pp. 719-729.
Pal, A. and Ceglarek, D. (2013), “Modeling of decision making process for product service failure
diagnosis”,Procedia CIRP, Vol. 11, pp. 32-37.
Pal, A., Franciosa, P. and Ceglarek, D. (2014), “Corrective actions of product service failures
via surrogate modelling of dimensional variations”, in Guan, Y. and Liao, H. (Eds),
Proceedings of the 2014 IIE Annual Conference Institute of Industrial Engineers-Publisher,
pp. 2271-2280.
Peng, Y., Dong, M. and Zuo, M.J. (2010), “Current status of machine prognostics in
condition-based maintenance: a review”,The International Journal of Advanced
Manufacturing Technology, Vol. 50 Nos 1/4, pp. 297-313.
Prakash, P.K.S. and Ceglarek, D. (2013), “Multi-step functional process adjustments to reduce
no-fault-found product failures in service caused by in-tolerance faults”,Procedia CIRP,
Vol. 11 No. 1, pp. 38-43.
Qian, Y., Xu, L., Li, X., Lin, L. and Kraslawski, A. (2008), “LUBRES: an expert system
development and implementation for real-time fault diagnosis of a lubricating oil refining
process”,Expert Systems with Applications, Vol. 35 No. 3, pp. 1252-1266.
Ruiz-Mezcua, B., Garcia-Crespo, A., Lopez-Cuadrado, J.L. and Gonzalez-Carrasco, I. (2011),
“An expert system development tool for non AI experts”,Expert Systems with Applications,
Vol. 38 No. 1, pp. 597-609.
Salfner, F. and Malek, M. (2007), “Using hidden semi-Markov models for effective online
failure prediction”,26th IEEE International Symposium on Reliable Distributed Systems,
SRDS 2007, IEEE, pp. 161-174.
1248
IMDS
115,7
Downloaded by NATIONAL TECHNICAL UNIVERSITY OF ATHENS At 04:29 29 July 2015 (PT)
Takata, S., Kimura, F., Van Houten, F.J.A.M. ,Westkamper, E., Shpitalni, M., Ceglarek, D. and
Lee, J. (2004), “Maintenance: changing role in life cycle management”,Annals of the CIRP,
Vol. 53 No. 2, pp. 643-656.
Tobon-Mejia, D.A., Medjaher, K. and Zerhouni, N. (2012a), “CNC machine tool's wear diagnostic
and prognostic by using dynamic Bayesian networks”,Mechanical Systems and Signal
Processing, Vol. 28 No. 1, pp. 167-182.
Tobon-Mejia, D.A., Medjaher, K., Zerhouni, N. and Tripot, G. (2012b), “A data-driven failure
prognostics method based on mixture of Gaussians hidden Markov models”,
IEEE Transactions on Reliability, Vol. 61 No. 2, pp. 491-503.
Veldman, J., Wortmann, H. and Klingenberg, W. (2011), “Typology of condition based
maintenance”,Journal of Quality in Maintenance Engineering, Vol. 17 No. 2, pp. 183-202.
Venkatasubramanian, V. (2005), “Prognostic and diagnostic monitoring of complex systems for
product lifecycle management: challenges and opportunities”,Computers & Chemical
Engineering, Vol. 29 No. 6, pp. 1253-1263.
Voisin, A., Levrat, E., Cocheteux, P. and Iung, B. (2010), “Generic prognosis model for proactive
maintenance decision support: application to pre-industrial e-maintenance test bed”,
Journal of Intelligent Manufacturing, Vol. 21 No. 2, pp. 177-193.
Waeyenbergh, G. and Pintelon, L. (2004), “Maintenance concept development: a case study”,
International Journal of Production Economics, Vol. 89 No. 3, pp. 395-405.
Wu, S.J., Gebraeel, N., Lawley, M.A. and Yih, Y. (2007), “A neural network integrated decision
support system for condition-based optimal predictive maintenance policy”,
IEEE Transactions on Systems, Man and Cybernetics, Part A: Systems and Humans,
Vol. 37 No. 2, pp. 226-236.
Further reading
Brotherton, T., Jahns, G., Jacobs, J. and Wroblewski, D. (2000), “Prognosis of faults in gas turbine
engines”,2000 IEEE Aerospace Conference Proceedings, IEEE, Vol. 6 No. 1, pp. 163-171.
Dumbill, E. (2012), Planning for Big Data,O’Reilly Media Inc., Sebastopol, CA.
Genc, E., Duffie, N. and Reinhart, G. (2014), “Event-based supply chain early warning system for
an adaptive production control”,Procedia CIRP, Vol. 19 No. 1, pp. 39-44.
Holmberg, K., Komonen, K., Oedewald, P., Peltonen, M., Reiman, T., Rouhiainen, V., Tervo, J. and
Heino, P. (2004), “Safety and reliability technology review”, VTT Industrial Systems,
Espoo.
Iung, B., Veron, M., Suhner, M. C. and Muller, A. (2005), “Integration of maintenance strategies
into prognosis process to decision-making aid on system operation”,CIRP Annals-
Manufacturing Technology, Vol. 54 No. 1, pp. 5-8.
Koc, M., Ni, J., Lee, J. and Bandyopadhyay, P. (2005), “Introduction of e-manufacturing”,
Proceedings of the 31st North American Manufacturing Research Conference (NAMRC),
Hamilton, pp. 97-1-97-9.
About the authors
Alexandros Bousdekis is a PhD Candidate and a Researcher at the Information Management
Unit, in the School of Electrical and Computer Engineering at the National Technical University
of Athens. He holds a diploma degree in production and management engineering from the
Technical University of Crete, Greece (2011) and a master of science in manufacturing systems
engineering from the Warwick Manufacturing Group (WMG) at the University of Warwick,
UK (2012). His research interests include event-driven computing, operational research and
decision making in manufacturing. Alexandros Bousdekis is the corresponding author and can
be contacted at: albous@mail.ntua.gr
1249
Proactive
decision
making
framework
Downloaded by NATIONAL TECHNICAL UNIVERSITY OF ATHENS At 04:29 29 July 2015 (PT)
Dr Babis Magoutas is a Senior Researcher at the National Technical University of Athens.
He holds a PhD in adaptive information systems (2010), an MBA in techno-economic systems
(2006) and a diploma degree in electrical and computer engineering (2003), all from NTUA.
During his studies he was awarded with two excellent performance scholarships from the Greek
State Scholarships Foundation and the Alexander S. Onassis Public Benefit Foundation.
His work focuses on event-driven, social and proactive computing, semantic web, knowledge
management, personalization and recommender systems. He has participated in more than eight
EU-funded IST projects, while he has published more than 22 papers in international
peer-reviewed journals and conferences in the areas of event-driven and proactive computing,
semantic web, personalization, information systems evaluation and collective intelligence. One of
his papers in the EGOV 2009 international conference received the “Best Paper Runner Up”
award in the category “the most interdisciplinary and innovative research contribution”. He has
been Programme Committee Member in the DEXA Workshop on Information Systems for
Situation Awareness and Situation Management, the International Workshop on Event-Driven
Business Process Management, the International Conference on Business Information Systems
and the International Conference on e-Business, while he has reviewed manuscripts for the
Internet Research Journal. He has also worked as a Software Engineer at Intracom SA.
Dimitris Apostolou is an Assistant Professor at the University of Piraeus, Greece and a Senior
Researcher at the Institute of Communication and Computer Systems, Greece. He holds
a PhD in knowledge management and decision support and his research concerns knowledge
management and knowledge-based decision support, group decision making, social- and event-
driven computing. He has professional experience in managing research and innovation ICT
projects and has participated in more than 20 projects funded by the European Commission.
He publishes in journals such as IEEE Intelligent Systems,International Journal of Information
Management,Expert Systems with Applications,Journal of Knowledge Management,Internet
Research. He is a member of the ΙΕΕΕ Computer Society.
Professor Gregoris Mentzas is a Full Professor of the Management Information Systems,
School of Electrical and Computer Engineering, National Technical University of Athens and
the Director of the Information Management Unit (IMU), a Multidisciplinary Research Unit at the
University. His area of expertise is information technology management and his research
concerns knowledge management, semantic web and social computing in e-government and
e-business settings. He has published four books and more than 200 papers in international peer-
reviewed journals and conferences, has two best papers awards in the ICE and e-Gov conferences,
sits on the editorial board of five international journals and has served as (co-)Chair or
Programme Committee Member in more than 55 international conferences. Gregoris has led or
contributed in 35 European research and development projects conducted in collaboration with
SAP, IBM, HP, Siemens, France Telecom, ATOS and other leading technology firms. Research
carried out by his group has led to the establishment of three internet technology companies.
He has acted as Grant Evaluator and/or External Reviewer in information technology programs
funded by donors such as the European Commission, the Swiss National Science Foundation, the
Austrian Science Fund and the Cyprus Research Promotion Foundation. His experience includes
12 years of management consulting in corporate strategy and information systems strategy.
He holds a diploma degree in engineering (1984) and a PhD in operations research and
information systems (1988) both from NTUA. During 2006-2009 he served as the Member of the
Board of Directors of the Institute of Communication and Computer Systems of NTUA.
For instructions on how to order reprints of this article, please visit our website:
www.emeraldgrouppublishing.com/licensing/reprints.htm
Or contact us for further details: permissions@emeraldinsight.com
1250
IMDS
115,7
Downloaded by NATIONAL TECHNICAL UNIVERSITY OF ATHENS At 04:29 29 July 2015 (PT)
