applied
sciences
Article
Techniques and Emerging Trends for State
of the Art Equipment Maintenance
Systems—A Bibliometric Analysis
Burkhard Hoppenstedt 1,*, Rüdiger Pryss 1, Birgit Stelzer 2, Fabian Meyer-Brötz 2,
Klaus Kammerer 1ID , Alexander Treß 3and Manfred Reichert 1
1Institute of Databases and Information Systems (DBIS), Ulm University, 89081 Ulm, Germany;
ruediger[email protected] (R.P.); klaus.kammer[email protected] (K.K.); manfr[email protected] (M.R.)
2Institute of Technology and Process Management, Ulm University, 89081 Ulm, Germany;
3ATR Software GmbH, 89231, Neu-Ulm, Germany; tress@atr-software.de
*Correspondence: burkhar[email protected]
Received: 28 March 2018; Accepted: 30 May 2018; Published: 2 June 2018
Abstract:
The increasing interconnection of machines in industrial production on one hand,
and the improved capabilities to store, retrieve, and analyze large amounts of data on the other,
offer promising perspectives for maintaining production machines. Recently, predictive maintenance
has gained increasing attention in the context of equipment maintenance systems. As opposed to
other approaches, predictive maintenance relies on machine behavior models, which offer several
advantages. In this highly interdisciplinary field, there is a lack of a literature review of relevant
research fields and realization techniques. To obtain a comprehensive overview on the state of the art,
large data sets of relevant literature need to be considered and, best case, be automatically partitioned
into relevant research fields. A proper methodology to obtain such an overview is the bibliometric
analysis method. In the presented work, we apply a bibliometric analysis to the field of equipment
maintenance systems. To be more precise, we analyzed clusters of identified literature with the goal to
obtain deeper insight into the related research fields. Moreover, cluster metrics reveal the importance
of a single paper and an investigation of the temporal cluster development indicates the evolution of
research topics. In this context, we introduce a new measure to compare results from different time
periods in an appropriate way. In turn, among others, this simplifies the analysis of topics, with a
vast amount of subtopics. Altogether, the obtained results particularly provide a comprehensive
overview of established techniques and emerging trends for equipment maintenance systems.
Keywords:
bibliometric analysis; predictive maintenance; research front graph; article density;
trend analysis
1. Motivation
This presented literature study systematically reviews existing literature related to equipment
maintenance systems to elaborate the state of the art as well as to discover current research trends.
In this context, predictive maintenance (PdM) constitutes a specific method that aims to improve
maintenance management methods, such as run-to-failure [
1
] or preventive maintenance [
2
], by including
knowledge of the machine behavior with the goal to derive an optimal maintenance strategy. According
to Mobley, the common target of predictive maintenance applications (PMA) constitutes “[...] regular
monitoring of the actual mechanical condition, operating efficiency, and other indicators of the
operating condition of machine-trains [...] to ensure the maximum interval between repairs as well as
to minimize the number and cost of unscheduled outages created by machine-train failures ” [3].
Appl. Sci. 2018,8, 918; doi:10.3390/app8060918 www.mdpi.com/journal/applsci
Appl. Sci. 2018,8, 918 2 of 29
In general, literature reviews [
4
] offer a technique to structure scientific articles. Existing review
types focus on different strategies and should be chosen to fit best to the problem definition.
Our requirements for reviewing the field of state of the art equipment maintenance systems are (1) to
cover a wide range of articles, (2) to identify contemporary maintenance procedures and techniques,
(3) to discover current trends, (4) to discover highly cited and influential articles, and (5) to include
latest research on equipment maintenance systems. Other literature reviews do not always satisfy
these requirements, e.g., the authors of [
5
,
6
] do not address Requirement 1. Concerning Requirement
2, for example, the authors of [
7
,
8
] do not consider recent research. In other works, in turn, the
analyzed equipment maintenance strategy is too specific [
9
,
10
] or too general [
11
] compared to the
literature review presented in this paper. To properly address Requirements 1–5, the literature review
we accomplished uses the bibliometric analysis. In particular, the bibliometric analysis offers several
advantages:
•
It visualizes relevant study articles based on various connection types, e.g., research fronts,
knowledge bases, authors, and author affiliations.
•It enables us to automatically cluster articles.
•It enables quick recalculations of clusters based on different time periods.
Consequently, bibliometric analyses tend to be very effective when facing large and heterogeneous
data sets. Thus, we apply the bibliometric analysis in the context of our literature review on equipment
maintenance systems. We believe that this research field involves a multitude of research issues as
well as complex interdependencies between them.
The procedure to accomplish the presented bibliometric analysis comprises ten steps as illustrated
in Figure 1in the left column. (Please note that that Figure 1also shows in the middle and right columns,
which steps have been actually accomplished for the present work.) The first step (cf. Figure 1, 1) is to
select the academic databases from which relevant works shall be obtained. Frequently, the following
databases are used: Google Scholar,Scopus,Web of Science, and ScienceDirect. Despite the wide-spread
use of Google Scholar in Computer Science, we decided to base our review on Web of Science for
two reasons. First, Web of Science and Scopus (Scopus would be another valuable source for our
approach. Please note that both Web of Science and Scopus are payment platforms [
12
]. The reason
we use just one is that our institution only has an account for Web of Science. In addition, according
to [
12
], the differences between the two platforms constitute no crucial threat to validity.) currently
generate citation data that can be comfortably used with existing bibliometric analysis software tools.
Second, Web of Science is compared to Google Scholar a so-called human-curated academic database,
which leads to more controlled citation data. On the other hand, Google Scholar reflects a more current
status of all published papers and theses in most cases. However, we still rely on Web of Science, as a
visual inspection revealed that only a few recent papers can be found with Google Scholar, but not in
Web of Science. Furthermore, an in-depth comparison of academic databases is a complex endeavor,
which is beyond the scope of this paper. However, we are aware of this limitation regarding the
presented review.
In the second step of a bibliometric analysis, one must specify the set of search strings to be
applied to the used academic database (cf. Figure 1, 2). In the context, we applied the following
search strings to Web of Science: predictive maintenance,condition monitoring,smart maintenance,
lean maintenance,reliability-centered maintenance, and predictive analytics. Those search strings were
defined after an open search on various maintenance techniques. All those techniques can be
also applied in the field of Engineering Asset Management ([
13
,
14
]). Concerning the 12 functions of
general asset management, as introduced by [
15
], we are mainly interested in Risk Management,
Condition Monitoring, Asset Usage Life Cycle, Performance Measures, and Information Systems. For
example, the integration of a maintenance method into asset management, e.g., by using condition
monitoring [16], represents one typical workflow that is relevant for us.
After applying the search strings to Web of Science, we exported the results (cf. Figure 1, 3) to
apply several preprocessing steps (cf. Figure 1, 4–5) to remove unfitting results (e.g., maintenance in a
Appl. Sci. 2018,8, 918 3 of 29
medical context) from the extracted data. Following this, the cleansed data was fed into a software for
bibliometric analysis. As depicted in Figure 1, the performed calculations and related interpretations
comprise five steps (i.e., Steps 5–10 in Figure 1). The obtained results, in turn, provide a valuable
overview for state of the art maintenance systems in general and predictive maintenance in particular.
Additionally, we were able to identify fundamental (i.e., distinguished) papers in this field. Moreover,
to the best of our knowledge, the work at hand is the first one that applied bibliometric analysis in
reviewing the research field of equipment maintenance systems. In addition, our results underlie that
bibliometric analyses are able to gain valuable insights.
The remainder of this paper is structured as follows: Section 2provides deeper insight into the
methodology applied, i.e., the bibliometric analysis. The result set obtained from the search strings
applied to Web of Science are discussed in Section 3. In Section 4, results of the bibliometric analysis are
presented. In turn, Section 5introduces a novel measure to rank single articles in a multi-dimensional
way. Section 6predicts research trends based on the past development of related topics, whereas
Section 7discusses threats to validity. Finally, Section 8concludes the paper and gives an outlook.
Select Data Source
1
Step ResultInput
Define Query String
2Comprehensive
Literature Study List of Search Queries
Export Articles from
Web of Science
3+List of Search
Queries
Data Sets
Calculated content Author content Data source
Select Fitting Categories
4Category Entries
from Data Sets
Whitelist of
Categories
Remove Unfitting
Categories
5Whitelist
Data Sets Filtered Data Sets
Calculate Clusters
6Filtered Data Sets Networks
Interpret Clusters
7Networks Research Fronts,
Techniques, Trends
Identify Top Articles
8Research Fronts List of Top Articles
Rank Top Articles
9Importance Measure
+ List of Top Articles
Sorted List
of Top Articles
Forecast
10 Trends Prognose
+
Figure 1. Workflow of a Bibliometric Analysis.
Appl. Sci. 2018,8, 918 4 of 29
2. Bibliometric Analysis
Bibliometric analysis is a method from the field of Bibliometrics [
17
]. The latter, in turn,
has connotations with the term statistical bibliography and constitutes a quantitative approach to
literature research. In the bibliometric analysis we performed with the goal to identify existing research
fields in the context of state of the art equipment maintenance systems, we follow the basic idea
presented in [
18
], which introduced mathematical graphs, denoted as networks, as a suitable solution
to represent a set of connected literature. In this network, each article is represented by a node, whereas
edges constitute their bibliographic connections. In turn, connections can be established based on a
multitude of similarity measures. In general, a similarity measure is based on the assumption that two
articles pose a similarity in terms of thematic content. To identify entire research fields, denoted as
clusters in a network (cf. Figure 2), two similarity measures are of utmost importance: Knowledge Bases
and Research Fronts [
19
] (cf. Figure 2). Two articles are bibliographically coupled (i.e., they are part of
a research front) if they reference the same third article. Couplings between nodes become stronger
(i.e., represented by a thicker line in the plotted network) if they co-cite multiple articles. By contrast,
two references are connected by co-citation (i.e., they are part of a knowledge base) if they are both
referenced by a third article. In addition to identify relevant research fields, we were interested in
networks for authors [
20
], countries, and organizations, which can be established by other similarity
measures. Based on the evaluation of networks, valuable results can be obtained according to [21]:
•Identifying the research field structure
•Detecting general contextual factors
•Discovering developments in research
•Forecasting coming trends
a3
a
1
a
2
a
3
a
1
a
2
co-citation
knowledge base research front
bibliographically coupled sample network
Cluster
Figure 2. Knowledge Base, Research Front, and Sample Network.
Another crucial aspect in performing a bibliometric analysis concerns the use of an appropriate
software tool. In the field of bibliometrics, there exist tools that are specifically developed for for
bibliometric analysis visualization techniques [
22
]). We used NetCulator [
23
] for the bibliometric
analysis we conducted. NetCulator offers features to calculate the mentioned networks. Furthermore,
it combines bibliographic coupling with lexical measures that, in turn, compare the similarity of
terms (cf. TF-IDF in Appendix A.2). Next, NetCulator calculates hybrid second-order similarities (See
Appendix A.2) with severe edge cutting [
24
]. The latter improves the calculation speed by ignoring all
weak edges that result from weak coupling between nodes. Finally, NetCulator computes clusters based
on Louvain Clustering [
25
]. Visualization and layouting for research fronts and knowledge bases either
rely on the OpenOrd-Method [
26
], which provides a quick overview by automatically summarizing
clusters, or the Fruchterman-Reingold algorithm [
27
] enabling a more detailed view. NetCulator offers
many other valuable features, we utilized the context of our bibliometic analysis, e.g.,
•Identifying core articles (i.e., significant papers within a research field).
Appl. Sci. 2018,8, 918 5 of 29
•Identifying frequently used terms in a research field.
•Identifying trends based on calculated timelines [28] for research fronts.
Although the technique of bibliometric analysis is promising, it also comes with some limitations:
First, the articles need to be published in an academic database. As a drawback, especially in the field
of industrial production, articles are often represented as patents and internal enterprise documents.
As another drawback, competition between companies leads to a large number of non-disclosure
agreements. Second, any bibliometric analysis is dependent on the choice of the used search strings [
29
].
Finally, a time lag must be considered, as it might take years for research to be published and cited.
3. Data Set
3.1. Search Strings
Based on a comprehensive literature prestudy, as well as interviews with domain
experts, we selected predictive maintenance [
3
], condition monitoring [
30
], smart maintenance [
31
],
lean maintenance [
32
], reliability-centered maintenance [
33
], and predictive analytics [
34
] as appropriate
search strings for our search strategy (cf. Figure 3).
Often applied
Brief Literature Review
We applied
Search String Strategies
Search Strings (s1 ... sn)
Apply Search Strings in
Academic Database
s1
s2 s3
s4
sn
Combine Search
Strings
Result Set
Apply Search Strings in
Academic Database
s1
Brief Literature Review
Search Strings (s1 ... sn)
Analyze Search
Strings for Categories
(c1 ... cn)
s2
s3 sn
filter categories
Result Set
Figure 3. Applied Search Strategy.
The Web of Science Core Collection is used to gather the article result set (date of withdrawal:
11 November 2016). Results for the search strings applied to Web of Science are shown in Table 1.
Appl. Sci. 2018,8, 918 6 of 29
Table 1. Overview of the gathered data set.
Query Entries Oldest Typical Category Articles from 2016
PDM a4742 1970 EEE * 311
CM b144,758 1956 EEE * 8416
SM c1447 1970 EEE * 109
LM d1299 1967 ND ** 74
RCM e588 1978 EEE * 27
PA f717 1996 EEE * 119
*
Engineering Electrical Electronic.
**
Nutrition Dietetics.
a
Predictive Maintenance.
b
Condition Monitoring.
c
Smart Maintenance. dLean Maintenance. eReliability-centered Maintenance. fPredictive Analytics.
In this paper, we particularly focus on the results related to predictive maintenance. Concerning
research history, the first article referring to predictive maintenance in its title [
35
] was published in
1970 in the field of chemical engineering. The number of papers published per year has significantly
increased since the 1990s (cf. Figure 4). For example, in 2015, the extracted predictive maintenance
articles had an average h-index [
36
] of 118 and a mean of 17.18 citations per item. By contrast, articles
dealing with condition monitoring have an average h-index of 36 and a mean of 2.97 citations per
item. Finally note that the search string condition monitoring is often used as a synonym for predictive
maintenance in case of a costly monitoring approach. Accordingly, the borders between predictive
maintenance and condition monitoring blur.
100
200
300
400
publications per year
1970
1980
1990
2000
2010
2016
publication year
Figure 4. Published Articles for Predictive Maintenance.
3.2. Preprocessing Dataset
As Web of Science categorizes each article based on a multitude of categories, first of all, we had to
choose the relevant ones. To simplify the categorization, we summarized Web of Science categories to
sections. Figure 5illustrates all sections (In addition, the left side of Figure 6illustrates the distribution
of sections for the search string predictive maintenance.) on one hand and the ones identified as being
relevant on the other. Based on the relevant Web of Science categories, we filter articles utilizing the
white list principle, i.e., we keep an article in the data set, if it matches at least with one of the relevant
categories. As can be obtained from Figure 5, we excluded a lot of Web of Science categories for the
search string predictive maintenance. First, we calculated an article density for sections (cf. Figure 6,
right side). The density is defined as (number of articles in section)/(number of categories in section). If the
density for a section is low, the section is excluded.
Appl. Sci. 2018,8, 918 7 of 29
Accepted
yesno
Section Categories (Selection)
Computer Science
Engineering
Management
Production
Environmental Science
Mathematics
Medicine
Physics
Social
Other
AI, Software Engineering, Hardware Architecture, Cybernetics
Electrical-, Mechanical-, Aerospace,- Nanotechnology
Operations Research, Business, Finance
Control Systems, Transportation, Industrial Engineering
Environmental Engineering, Biodiversity Conservation
Statistics, Applied Mathematics
Surgery, Immunology, Health Care
Optics, Thermodynamics, Nuclear
Education, Political Science, Sociology
Sport, Food Science
Figure 5. Relevant Web of Science Categories.
For example, in the section medicine (cf. Figure 6), the density is very low, meaning that the
average number of articles per category is small. The reason we exclude such sections is based on the
assumption that predictive maintenance solely plays a side role in these sections.
Articles per Subcategory Number of Subcategories
(* density = Articles per Subcategory / Number of Subcategories)
Figure 6. Section Distribution and Article Density (Improved Readability).
4. Biblometric Analysis Results
We apply the bibliometric analysis to the data set described in Section 3using the NetCulator tool.
Thereby, for each aspect to be evaluated (i.e., research fronts, knowledge bases, authors, organizations,
countries, terms), several NetCulator parameters have to be defined.
First of all, a time period needs to be specified. For our bibliometric analysis, we decided
to do the calculations for different time periods in order to identify how the topic of predictive
maintenance evolves over time. More specifically, we use three time periods: [1990–2000], [2000–2008],
and [2008–2016]. These periods were defined using an iterative process in NetCulator and comparing
emerging paper distributions. Please note that that this is a common way pursued in a bibliometric
analysis. Then, the NetCulator user must define many more parameters (e.g., cluster size). Thereby,
the user can either use default parameters or specifies his own parameter set. All parameters that can
be specified are shown in Figure 7. Finally, NetCulator presents the calculated networks to the user.
Please note that that the Figure 7shows the actual parameters used for the presented bibliometric
analysis in this work.
Appl. Sci. 2018,8, 918 8 of 29
Figure 7. Calculation Parameters (Improved Readability).
4.1. Predictive Maintenance
In the following, we provide a detailed analyses for predictive maintenance in the aforementioned
time periods (cf. Figure 8).
PdM Period 1 PdM Period 2 PdM Period 3 CM Related Methods
Country Network
Keyword Ranking
Research Front Clusters
Research Front Graph
Publication Map
Research Front Flow Diagram
Keyword Ranking Comparison
Coorporation Map
Identified Techniques
3D Density Matrix
Moved to Appendix
Figure 8. Visualizations in each Period.
4.1.1. Period 1990–2000
First of all, the results for the first period are based on 175 articles. The first work for this period
was initiated by the American Association for Artificial Intelligence, whose subgroup, called Special
Interest in Manufacturing (SIGMAN), organized workshops collecting ideas in this context. Since an
American organization was pioneering the topic of predictive maintenance in this period, the leading
publication nation was the US (29.7%), followed by France (8.0%) and Canada (7.4%). According to
Figure 9, we can observe an increasing cooperation of countries over time and a stronger thematic
focus. More specifically, each data point in Figure 9represents a country, a close distance between them
indicates researching in similar topics. However, closeness does not necessarily display collaborations,
but constitutes an indicator of a high knowledge transfer between the involved countries. In general,
the research in this first period did not reveal any outstanding research topic. Accordingly, no leading
author or institution could be identified. The keyword analysis for the first period (cf. Table 2) shows a
strong connection between the terms predictive maintenance and monitoring as they belong to the same
cluster (i.e., Cluster 1from Table 2. Furthermore, as monitoring is represented by the two keywords
condition monitoring and monitoring, this topic is a leading one in this period.
Appl. Sci. 2018,8, 918 9 of 29
1990 - 2000 2000 - 2008 2008 - 2016
Publication density Countries Strong cooporation
Figure 9. Evolution of Country Networks (obsolete).
Interestingly, monitoring and predictive maintenance are in the same cluster, while this does not
apply to condition monitoring, i.e., in this period, despite the thematic closeness of condition monitoring
and predictive maintenance, they were addressed independently. Please note that that this result supports
our decision to consider condition monitoring as a separate search string. In general, we assume the
following order of complexity for tasks in the context of equipment maintenance systems:
Report →Analysis →Monitoring →Prediction
Furthermore, we assume the technique monitoring needs mainly humans as decision makers and
rarely relies on techniques of machine learning.
Table 2. Keywords.
Keyword Quantity Percentage Cluster
Predictive Maintenance 25 14.3 1
Condition Monitoring 12 6.9 3
Maintenance 8 4.6 1
Diagnosis 7 4.0 2
Monitoring 6 3.4 1
Models 6 3.4 1
Neural Networks 5 2.9 2
Vibration 5 2.9 2
After calculating research fronts, we sort the articles based on the number of citations.
The most popular paper, according to citations, is denoted as lead document. In a manual inspection,
used techniques of the lead document and a thematic field for each cluster have been identified.
The collected information are shown in Table 3, where the thematic field is denoted as cluster label.
The manual labeling of each cluster, denoted as cluster caption, and the identification of the used
technique considers top terms, scanning the resulting articles. Please note that not every article in each
research front needs to deal with the topic the cluster caption implies. In general, a bigger cluster size
increases the probability of a false-positive cluster classification. The research front clusters from Table 3
are explained in the following. Cluster 1 deals with the continuous status monitoring. This cluster is
coined by a joint project between the US and Egypt, as represented by the Alexandria National Iron &
Steel Company. Second, in Cluster 2, items were found dealing with vibration analysis. Here, vibrations
caused by for example pumps, spindles and motors are used to refer to abnormal machine behavior.
Cluster 3 assembles literature that deals with maintenance in risky environments, such as oil refinery
and nuclear power plants. With a focus on special electronic equipment, such as radars and motors,
Cluster 4 contains research based on the question how to predict electric systems behavior.
Appl. Sci. 2018,8, 918 10 of 29
Table 3. Research Front Clusters.
# Lead Document Citations Cluster Label Technique
1Predictable scheduling of 173 On-line Monitoring Network Minimax
a job shop subject to breakdowns [37]
2 Analysis of computed order tracking [38] 173 Vibration Analysis Frequency analysis
3The analytic hierarchy process applied 133 Nuclear Risks and Safety Analytic Hierarchy Process
to maintenance strategy selection [39]
4Insulation failure prediction in AC 82 Prediction on Electric Systems Band Pass Filter
machines using line-neutral voltages [40]
5Pontis: A System for Maintenance Optimization 53 Stochastic Models Markov Processes
and Improvement of US Bridge Networks [41]
6Classification techniques for metric-based 50 Software Maintenance Fuzzy Classification
software development [42]
7Model-predictive control of a combined 39 Model Predictive Control Model Predictive Control
sewer system [43]
8Emissivity measurement and temperature 26 Emission Control IR Camera
correction accuracy considerations [44]
9Prediction of wafer state after plasma processing 23 Hydroelectricity & Photovoltaics Neural Network
using real-time tool data [45]
Cluster 5 represents the research field of stochastic approaches in various applications. Since not
only machines need to be supervised, but also the software itself, the field of software maintenance 6
addresses these challenges with code metrics and fuzzy structures. Cluster 7 lists literature dealing
with model predictive control (MPC). It is an advanced method of process control and it is frequently
used in the context of lime production and kiln heating. Next, literature concerning emission control
is collected in Cluster 8. Emission changes, e.g., caught by a infrared camera, indicate a changing
machine behavior. Finally, Cluster 9 is labeled with the term hydroelectricity & photovoltaics. In this
cluster, agent-based methods can be frequently found.
4.1.2. Period 2000–2008
First, research fronts and lead documents are extracted analogous to the way described in
Section 4.1.1 (cf. Table A1). Some research front labels from the previous period can be reused, while
new clusters are presented in the following. The topic of predictive models (1) bundles general approaches
to model and improve maintenance approaches, including even basic ideas (e.g., corporations between
firms for data exchange). In the previous period, the keyword analysis recognized monitoring (3) as
being important, but no further cluster were identified for monitoring. In turn, during the second
period, there exists a cluster with a focus on visualization. Third, when considering maintenance,
costs constitute an important factor as well. The cluster of cost-based models (4) emphasizes the costs
for exchanging and maintaining machine parts. This cluster can be seen in contrast to the approach
of reliability-centered maintenance. Concerning abrasion and the condition of cables, the cluster cable
diagnostic (8) discusses solutions. With growing complexity of maintenance systems, the management
(9) aspect is evolving accordingly. In this cluster, the literature stresses the organization of maintenance
strategies as well as the evaluation of their performance. Next, research on motor fault prediction (10)
grew forming a cluster. A widely used method in this context is the root cause analysis. Moreover,
a new cluster denoted as generic approach (11) was found. Thereby, in most papers data-mining or
machine-learning methods are utilized to enable adaptive maintenance systems. Finally, the cluster
sensor data (13) deals with the integration of sensors into maintenance systems.
A visualization of the research fronts corresponding to Table A1 is illustrated in Figure 10. Articles
from the same cluster are plotted with the same shape and color, whereas their proximity indicates a
thematic similarity. Please note that a high article density is marked with dark blue. Figure 10 depicts
all 574 papers from the second period. When regarding their connections, however, only the strongest
242 out of 4826 are shown. The research front managing pdm is the central cluster, as it is an indirect
part of many other research fields. By contrast, the software prediction cluster is outside of the center as
it is not thematically connected to most other clusters.
Appl. Sci. 2018,8, 918 11 of 29
motor fault prediction
vibration analysis
software prediction
model predictive control
cable diagnostic
thermography
scheduling model
managing pdm
generic approach
Figure 10. Research Front Graph.
Concerning participating countries in research, the increase of Chinese publications is noticeable.
More specifically, the number of publications grew from 0 to 54 (cf. Figure 11), making China the
second most publishing country in this field. Remarkable seems to be the fairly low participation in
research of highly industrialized countries like Japan and Germany. The United States remain the most
active country in the research field of predictive maintenance and the University Cincinnati (Ohio)
constitutes the driving force.
4.1.3. Period 2008–2016
Again, first of all, research fronts are calculated and evaluated (cf. Table A2). The newly emerging
clusters are presented in the following. First, one of the added clusters deals with multi-component
applications (1), that aim to combine data of multiple production units into one predictive model.
Appl. Sci. 2018,8, 918 12 of 29
2 377
Number of Publications
Figure 11. Published Literature.
With increasing interest in alternative energies, wind turbines (4) come into the focus of predictive
maintenance systems. Third, the discipline known as prognostics and health management (5), which links
failure mechanisms to system life cycle management, forms a cluster as well. Furthermore cluster Cloud
and IoT (6) aims at better production results by, first and foremost, including recent developments from
Big Data Analytics and Distributed Systems. Dealing with certain machine components, the analyzes of
rolling bearings (7) and capacitors (8) form other clusters. Literature referring to the prediction of metal
deterioration due to corrosion is collected in (11), whereas rail maintenance (14) monitors the current state
of roads and rails. Finally, by analyzing changes in oil (15) consistency and consumption, a changing
machine behavior must be considered. Concerning the publishing countries, China increases its
number of publications, while the US remain number one. In turn, the Tsinghua University (Peking)
constitutes now the driving force for publications in this topic.
After discussing the results of the periods separately, keywords, research fronts, and techniques
over all three periods are juxtaposed in the following.
Keyword Ranking
A first possibility of showing the development of research during the time periods is the
comparison of top keywords. Table 4shows the top mentioned key words, their percentage distribution,
and their difference in ranking position compared to the previous period (cf. Section 4.1.1). The ranking
difference is denoted as
∆
, whereby a positive
∆
denotes growing importance of this keyword.
Furthermore, letter nstands for not mentioned yet, meaning that this keyword was not introduced in
a previous period. Two particular developments can be obtained from Table 4. First, the complexity
of predictive maintenance systems rises over time. During the first period, there is a focus on rather
simple methods, such as monitoring and preventive maintenance, while their importance decrease
during the last period, in which more complex issues (e.g., prediction and prognostics) are addressed.
An example of this development can be seen in the ratio between fault detection and fault diagnosis.
During the first period, they are equally important, while in the last period the diagnostic part is more
important. As a second aspect, the used techniques are not explicitly mentioned anymore in the last
period. The disappearance of techniques like neural networks and genetic algorithms, which had an
outstanding ranking position before, indicates that they are now in a rather major state, i.e., they are
no longer explicitly mentioned.
Appl. Sci. 2018,8, 918 13 of 29
Table 4. Keyword ranking.
1990–2000 2000–2008 2008-2016
# Key % Key % ∆Key % ∆
1 predictive maintenance 17.99 predictive maintenance 11.86 0 predictive maintenance 9.61 0
2 condition monitoring 8.63 maintenance 3.83 +1 condition monitoring 3.17 +1
3 maintenance 5.76 condition monitoring 3.35 −1 maintenance 2.73 −1
4 diagnosis 5.04 neural network 2.75 +3 reliability 2.12 +1
5 monitoring 4.32 reliability 2.51 +27 condition-based maintenance 1.5 +7
6 models 4.32 fault detection 1.92 +7 fault diagnosis 1.32 +11
7 neural networks 3.6 preventive maintenance 1.8 +2 prognostics 1.28 +2
8 vibration 3.6 genetic algorithm 1.56 n prediction 1.15 +36
9 preventive maintenance 2.88 prognostics 1.44 n model predictive control 0.93 +67
10 emissivity 2.16 partial discharges 1.32 n data mining 0.93 +8
11 thermography 2.16 condition-based maintenance 1.32 n preventive maintenance 0.93 −4
12 fault diagnosis 2.16 diagnostics 1.2 +22 fault detection 0.88 −6
13 fault detection 2.16 modelling 1.08 n optimization 0.79 +48
14 failure prediction 2.16 sensors 0.96 n wind turbine 0.75 n
15 failures 1.44 maintenance management 0.96 n monitoring 0.71 +7
Research Front Development
To gain insights into the development of research fronts, we sort all identified research fronts
according to their relative size in the period and search for recurrences (cf. Figure 12). Furthermore,
we identify highly overlapping subjects, using the graphical representation of research fronts, and
put related research fronts in brackets. As vibration analysis is constantly present in all periods, it is
being revealed as the most important research field. In addition, we discovered that the thematic field
Internet of Things gains importance over time and it is connected to generic approaches as well.
Prediction in Electric Systems
Emission Control
Hydroelectricity
Model Predictive Control
Vibration Analysis
Software Maintenance
On-line Monitoring Network
4% > clustersize
4% ≤ clustersize < 8%
8% ≤ clustersize < 12%
12% ≤ clustersize < 16%
16% ≤ clustersize
Stochastic Models
Nuclear Risks and Safety
Vibration Analysis
Predictive Models
Motor Fault Prediction
Wind Turbines
Generic Approach
Sensor Data
Managing PdM
Monitoring
Software Prediction
Cable Diagnostics
Thermography
Model Predictive Control
Cost Based Model
Cloud and IoT
Prediction in Electric Systems
Prognostics and Health Management
Model Predictive Control
Rolling Bearing
Capacitors
Corrosion
Multi-Component Systems
Rail Maintenance
Oils
Vibration Analysis
Generic Approach
Software Prediction
Thermography
Nuclear Risks and Safety
1990-2000 2000-2008 2008-2016
Figure 12. Research Front Flow Diagram (Color and Font Sizes).
Technique Identification
A final comparison over all three time periods aims to identify the techniques applied in the
maintenance approaches. Thereby, for the lead document of each cluster, the used technique is extracted
from the abstract and keywords as well as by evaluating the context of the document. In addition,
the identified techniques were manually classified into the following four categories:
•Technical. The approach focuses on application-specific technical parameters.
•Stochastic. The approach emphasizes the importance of stochastic issues.
•Generic. The approach deals with adaptive and learning methods.
•Process. The approach focuses on the maintenance lifecycle.
Appl. Sci. 2018,8, 918 14 of 29
Based on these categories, the identified techniques are visualized in a timeline (cf. Figure 13)
to obtain further insights. In general, we can observe a dominance of technical approaches. Thereby,
most research focuses on concrete machine use cases, whereas general frameworks are less considered.
1990 20001995 2005 2010 2015
Technical Stochastic
Band Pass Filter
IR Camera
Neural Network
Markov Process
MPC
Generic
Frequency Analysis
Fuzzy Classification
Process
AHP
Minimax
SOM
Interfirm Nets
RCA
FEA
HTC
Analog Signature Analysis
Total Productive Maintenance
e-Maintenance
ROSE Framework
FA
IR Camera
MCO
Data-mining
CSM
Decision Framework
PIDC
Genetic Algorithm
Kalman Filtering
Corrosion Rate
Markov Process
Degradation Model
Fuller‘s Earth
Dissipation Factor
SVM
MPC Model Predictive Control
RCA Root Cause Analysis
FEA Finite Element Analysis
FA Frequency Analysis
HTC Heat Transfer Coefficient
CSM Covariance Structure Model
AHP Analytic Hierachy Process
IR Infrared
MCO Maintenance Cost Optimization
SOM Self Organizing Map
SVM Support Vector Machine
PID Proportional-Integral-Derivative Controller
Figure 13. Timeline of Techniques.
4.2. Predictive Analytics, Reliability-Centered Maintenance, Predictive Analytics, and Smart Maintenance
We combine the datasets of four search strings in order to obtain cluster sizes comparable to the
ones for search strings predictive maintenance and condition monitoring. The resulting dataset, in turn,
contains 2330 items. Concerning countries, similar to the results we obtained for predictive maintenance,
the US is the driving force in this research field, with the Universities of Washington, Illinois and
Cincinnati being leading organizations for the time between 1990 and 2016. Please note that the
University of Cincinnati was another driving force in the field of predictive maintenance in the period
from 2000 to 2008. Finally, Figure 14 reveals corporations of countries from 2008 to 2016.
Appl. Sci. 2018,8, 918 15 of 29
cluster
1
2
3
4
Research clusters
Figure 14. Cross-Country Cooperations during 2008–2016.
4.2.1. Period 1990–2000
Analogous to the approach used for predictive maintenance, research fronts are calculated and
evaluated (cf. Table 5). If a research field for this evaluated search string was also identified in
Section 4.1, it gets marked with (1)–(3). The numbers (1)–(3) represent periods. For example, the research
field sensors from Table 5was found in the second period of predictive maintenance, with the following
results obtained: First, lean networks (1) deal with the process optimization for maintenance, especially
for companies that want apply the method of lean production.Reliability-centered maintenance, in turn,
can be frequently found in aircraft (2) applications due to the high safety standards required in this
field. Finally, explanations on nuclear power plants (3) and sensors (4) can be already found in Section 4.1.
Table 5. Research Front Clusters.
# Lead Document Citations Cluster Label Technique
1 Collaborative Advantage—The art of alliances[46] 274 Lean networks Alliances
2 Reliability centered maintenance [47] 92 Aircraft RCM
3Optimization of standby safety system maintenance 76 Nuclear Power Plants (1),(3)Dynamic Programming
schedules in nuclear power plants [48]
4Stochastic crack growth analysis methodologies 36 Sensors (2)Crack Growth Analysis
for metallic structures [49]
4.2.2. Period 2000–2008
For the second period, the research fronts are shown in a 3D visualization (cf. Figure 15). Figure 15
is based on the same calculated data as Figure 10 (cf. Section 4.1). However, the article density, denoted
as
ρ
, is represented through the height of the 3D surface plot. Thereby, local peaks in the surface
plot represent research fields. The latter, in turn, are partly aggregated to finally form seven research
fronts of this period (cf. Table 6)). Please note that the results of the search strings reliability centered
maintenance and smart maintenance can be clearly separated.
Appl. Sci. 2018,8, 918 16 of 29
Smart Maintenance
Reliability Centered Maintenance
y
x
ρ(x,y)
Figure 15. Article Density Surface.
The first cluster revealed is denoted as lean manufacturing (1). It is strongly connected to cluster
lean maintenance discovered in the previous period. Second, a main cluster deals with maintainable
systems (2). Process-oriented approaches, in turn, are collected in cluster lifecycle management (3).
Moreover, clusters condition-based maintenance (4) and reliability-centered maintenance (5) are considered.
Finally, smart networks (7) often integrate smart sensors in the production lifecycle.
Table 6. Research Front Clusters.
# Lead Document Citations Cluster Label Technique
1bundles, and performance [50]475 Lean Manufacturing Lean Manufacturing
bundles, and performance [50]
2The present status of maintenance strategies 116 Maintainable Systems Maintainnable System
and the impact of maintenance on reliability [51]
3Research issues on product lifecycle management 86 Lifecycle management Lifecycle Management
and information tracking using smart embedded systems [52]
4System health monitoring and prognostics—a 85 Condition-based Health Monitoring
review of current paradigms and practices [53] Maintenance
5Reliability-centered predictive maintenance scheduling 82 Reliability-centered Reliability-centered
for a continuously monitored system subject to degradation [54] Maintenance Maintenance
6Aircraft composites assessment by means 71 Aircraft Non-destructive Testing
of transient thermal NDT [55]
7Wireless Industrial Monitoring and Control 52 Smart Networks Bluetooth
Using a Smart Sensor Platform [56]
4.2.3. Period 2008–2016
Table 7shows the results for this period. Smart meters (1) constitute the first newly discovered
research field. These intelligent electronic devices enable two-way communication between meter
and production system. Furthermore, Big Data Analysis (3) and Data Science (4) are focusing on data
analytics. The increasing capabilities of distributed systems increase the potential for automation systems
(4). Finally, total productive maintenance (7) constitutes a method including key performance indicators
(KPI) for production control.
Appl. Sci. 2018,8, 918 17 of 29
Table 7. Research Front Clusters.
# Lead Document Citations Cluster Label Technique
1Smart meters for power grid: Challenges,
issues, advantages and status [57]87 Smart meters Power Grid
2
Prognostics and health management design
for rotary machinery systems-Reviews,
methodology and applications [58]
66 Prognostics and Health
Management (3)
Prognostics and
Health Management
3
Predictive analytics in information systems
research [59]52 Big Data Analysis Statistical Model
4
Service-Oriented Infrastructure to Support
the Deployment of Evolvable Production
Systems [60]
47 Automation Systems Web Services
5
Data Science, Predictive Analytics, and Big
Data: A Revolution That Will Transform
Supply Chain Design and Management [
61
]
46 Data Science Supply Chain
management
6
Development of an optimized
condition-based maintenance system by
data fusion and reliability-centered
maintenance [62]
44 Reliability-centered
Maintenance Data Fusion
7In pursuit of implementation patterns: The
context of Lean and Six Sigma [63]44 Total Productive
Maintenance
Total Productive
Maintenance
8
Reliability-Centered Maintenance for Wind
Turbines Based on Statistical Analysis and
Practical Experience [64]
39 Wind Turbine (3)Failure Mode and
Effect Analysis
9Robust Self-Healing Concrete for
Sustainable Infrastructure [65]39 Smart Material Self-healing Concrete
10
Intelligent Systems for Improved Reliability
and Failure Diagnosis in Distribution
Systems [66]
28 Smart Grid Distribution Fault
Anticipation
11
A survey on virtual machine migration and
server consolidation frameworks for cloud
data centers [67]
27 Cloud Computing (3)VM Migration
12
State of the art review of inspection
technologies for condition assessment of
water pipes [68]
26 Road Maintenance Non-destructive
Inspection
13 Direct Evaluation of IEC 61850-9-2 Process
Bus Network Performance [69]12 Circuit Breakers Smart Grids
14 Electrical energy storage systems:
A comparative life cycle cost analysis [70]11 Energy Systems Monte Carlo
4.3. Condition Monitoring
Papers related to condition monitoring contribute the largest dataset. To allow for a comparability
with the other data sets, calculation parameters have to be adapted. To be more precise, the average
cluster size is increased to 200, whereas each period of time is limited to one year. Hence, the
periods differ from the previously defined ones. Following this approach, twelve research fronts were
calculated.
Table 8lists the discovered lead documents and techniques for each year. Interestingly, many
known research fields from previous Sections could be observed and only two new research fields
were discovered at the end: First, data-driven systems (8) enrich production workflows by including
data evaluation methods. Second, cyber-physical systems (12) focus on the combination of physical and
software components.
4.4. Engineering Asset Management
As we also consider that the revealed maintenance methods should be applicable in the field of
Engineering Asset Management, which is not used as a direct search string, we discuss this connection
briefly. Therefore, we show connections between revealed papers and this research field using the
definition of general asset management functions as presented by the authors in [
15
]. First, Risk
Management is the process of identifying risks and taking steps to reduce the latter [
71
]. In case of
equipment maintenance, this would be the identification of the Remaining Useful Life (RUL) [
72
]. It
is evident that Condition Monitoring can be directly found in our revealed papers, e.g., as shown by
[
73
]. Next, the Asset Usage Life Cycle is of paramount importance and [
70
] uses this concept for energy
storage systems. Moreover, maintenance strategies have a great impact on Performance Measures, as
Appl. Sci. 2018,8, 918 18 of 29
shown, for example, by the authors of [
74
]. Finally, Intelligent Information Systems enrich maintenance
approaches by improving the reliability [
66
]. In conclusion, our revealed papers cover many parts of
the Engineering Asset Management field.
Table 8. Research field and technique of the most cited papers or corresponding clusters 2005–2012.
# Year Lead Document Citations Cluster Label Technique
1 2005 [75] 591 Motor Analysis (2)Motor Current Signature Analysis
2 2006 [76] 936 Prognostics and Health Management (3)Sensor Data Fusion
3 2007 [77] 284 Motor analysis (2)Support Vector Machine
4 2008 [78] 187 Motor analysis (2)Principle Component Analysis
5 2009 [79] 297 Wind Turbines (3)Fault Detection System
6 2010 [80] 375 Sensor-based Systems (2)Wearable Systems
7 2011 [72] 248 Monitoring (2)Remaining Useful Life
8 2012 [81] 380 Data-driven Systems Process Monitoring
9 2013 [82] 90 Thermography (2),(3)IR-Camera
10 2014 [83] 243 Monitoring (2)Data-Driven Approach
11 2015 [84] 243 Monitoring (2)Data-Based Techniques
12 2016 [85] 1 Cyber-physical Systems Adaptive Learning
5. Importance Measure
By collecting all lead documents for each time period, we obtain lists with potentially relevant
papers. Please note that the importance of literature entries in these lists is not directly comparable
for two reasons. First of all, when judging the importance of a paper, not only the number of its
citations should be considered. Additionally, for two papers with same number of citations, their date
of publication should be considered as well. For this purpose, we introduce a correction factor
tcn
(cf. Equation (2)). Second, the citing culture differs across the various research fields. For example,
in economics, the mean number of citations is higher than in computer science. In line with the latter
observation, regarding the results of search string Lean Maintenance, we obtain articles with high
numbers of citations that mainly focus on management-oriented issues ([
46
,
50
]). To reduce these two
effects and to express aspects from the bibliometric analysis more properly, we propose an alternative
impact measure. (For a general overview of impact measures see [86]).
The first part of the importance measure represents anomalies caused by the different search
strings (Predictive Maintenance (PDM), Condition Monitoring (CM), Predictive Analytics (PA),
Reliability-centered Maintenance (RCM), Lean Maintenance (LM), and Smart Maintenance (SM)).
Assuming that the search strings differ in their relevance for state of the art equipment maintenance
systems, we express their importance on scale from 0 to 1. This factor is denoted as topic neighborhood
(tn) assigned to each article a:
tn(a) =
1.0, a∈ {PDM}
0.7, a∈ {CM}
0.4, a∈ {PA ∪RCM ∪LM ∪SM}
(1)
As opposed to the h-index [
36
], which cannot decrease over time, we apply a logarithmic reduction
(cf . Figure 16) for the value of a citation. For example, Figure 16 shows a decreasing, normalized
influence value for 88 fictive citations over a time span of 18 years.
Appl. Sci. 2018,8, 918 19 of 29
tn = 1
tn = 0,7
tn = 0,4
# years
0 2 4 6 108 12 1614 18
importance
0.1
0.2
0.3
0.5
0.4
0.6
0.7
0.8
0.9
1.0
0.2
Figure 16. Logarithmic Importance Reduction.
Since the bibliometric analysis is based on clustering, it seems to be appropriate to include the
relative cluster size (cs) of a cluster cin period pas well. In addition, the average h-index of the
first and last author will be part of the importance measure to express that literature from authors
widely read is very important. Finally, graph-based indicators contribute to the importance measure.
The important papers from a cluster can be identified using the article degree, also named as well
valency [
87
]. Besides, according to the graph theory, a cluster can be denoted as a graph G, with each
article representing a vertex v, where deg(v) is the number of edges incident to the vertex. A second
graph unit is betweenness [
88
], which constitutes a centrality measure. Betweenness represents the
number of shortest paths in the cluster that cross v. Clusters containing a high number of items
automatically produce large betweenness values and. Hence, the betweenness needs to be calculated
relatively to the cluster size.
In a nutshell, the importance index of an article, suggested by us, considers the
following attributes:
•age t[2016-publish year]
•times cited tc
•cluster size cs
•number of items npin the assigned period p
•h-index mean (hi) of the first and last author
•topic neighbourhood tn
•degree d
•betweenness b
The newly introduced importance measure refers to five different variables (cf. Equation (2)),
whereby each term is weighted according to the preferences of the study leader. For our analysis,
we stress the importance of the centrality measures. (cf. Equation (3)).
tcn=
tc ·tn
(log(t+1) + 1)
,csn=
100 ·cs
np
,hn=
hi
,dn=
d
,bn=
b
np
(2)
The resulting impact factor for each article is then calculated as follows:
im =0.3 ·tcn+0.15 ·csn+0.05 ·hn+0.25 ·dn+0.25 ·bn(3)
The impact measures of selected articles and the parameters involved are shown in Table 9(all
values are normalized), accompanied by the statistical measures average (avg), median (med), and
standard deviation (std). Please note that the high weight of the centrality measures leads to top positions
for articles showing excellent degree and betweenness values. Exemplarly, we will discuss the first
Appl. Sci. 2018,8, 918 20 of 29
row of Table 9. The paper was cited 53 times and was 5 years old when the collection process took
place. It is indirectly connected to maintenances and receives therefore the topic neighbourhood of 0.4.
According to Equation (2), the value
tcn
is calculated and is 0.03. This number indicates that this paper
is relatively seldom cited among all papers. Next, the paper was revealed as the most important paper
in a cluster of size 318 in a period with 1813 entries. This results in a relative cluster size of about 17%
and a normalized value of 0.43. According to Web of Science the h-Index of the first and last author
are 17 and 15, which produces a normalzed value of 0.4. Both degree and degreeness are the highest
values among all papers and therefore represented by the normalized value 1. Using Equation (3),
the final importance measure for this paper is 0.59.
Table 9. Article Ranking Sorted by Importance Measure
# Cite Times Cited Cluster Size H-Index Degree Betweenness Importance
1 [59] 0.03 0.43 0.40 1 1 0.59
2 [89] 0.28 0.05 0.43 0.28 1 0.43
3 [76] 1.00 0.18 0.36 0.18 0.02 0.40
4 [38] 0.07 0.32 0.33 0.18 0.90 0.36
5 [75] 0.62 0.43 0.26 0.32 0.02 0.35
31 [47] 0.89 0.15 0.02 0.31 0.06 0.14
32 [51] 0.05 0.43 0.38 0.15 0.00 0.14
73 [69] 0.01 0.03 0.31 0.14 0.00 0.06
74 [67] 0.02 0.03 0.02 0.14 0.00 0.05
avg 0.1 0.21 0.25 0.21 0.10 0.15
med 0.05 0.18 0.19 0.17 0.03 0.12
std 0.16 0.16 0.21 0.12 0.22 0.10
Finally, two other aspects need to be mentioned. First, the presented ranking can be easily
parameterized by changing the values of tn and the used weights. Second, the introduced importance
measure is well suited for a complex bibliometric analysis including various research areas.
6. Forecast
We forecast the development of identified research fronts based on the history of scientific and
public interest. According to [
90
], scanning constitutes a prerequisite of forecasting. Scanning involves
“collecting information: the system, history, and context of the issue [...]”. We collected the required
information from our bibliometric analysis. For this forecast, we only include those research fields
connected to search string predictive maintenance. Thereby, we assume that a higher interest in the topic
will result in more publications. Furthermore, we distinguish between scientific and public interest.
To describe the scientific interest, six research fronts from the period 2008–2016 were chosen and their
names were used as new search queries for Web of Science. This period constitutes the latest one and,
hence, the most important period for us. From this period, we take six important research fronts and
analyze their development over the last 20 years; i.e., according to the published items per year. . This
time, we were not interested in extracting literature, but in measuring the published items per year (cf.
Figure 17). Interestingly, the development over the last 20 years has shown a significant growth in
almost all connected research fields. In particular, the research front Internet of Things is identified to be
a potential driving force for publications in the context of predictive maintenance.
Appl. Sci. 2018,8, 918 21 of 29
IoT Vibration Analysis
ThermographySoftware Maintenance
Wind turbines
Nuclear Plants
1995-2015 1995-2015 1995-2015
published items per year published items per year
1000
1000
Figure 17. Published Items Per Year.
As an indicator for public interest, we choose the Google trend analysis. (https://www.google.
de/trends/explore) Interest, as measured by Google searches, is visualized in a normalized form for
the period from 2004 to 2017 (cf. Figure 18). The topic Internet of Things grew significantly during the
last years. The greatest difference between scientific and public interest can be observed in the graphs
related to software maintenance. Please note that the Fukushima Daiichi nuclear disaster in March 2001
distorts the graph for nuclear power plants. The public interest was enormous due to this event, causing
a peak in the graph.
Wind
turbines
Internet of
Things
Vibration
Analysis
Software
Maintenance
Thermography
Nuclear
Plants
Jan 2004 Jan 2017
100 %
0 %
100 %
0 %
100 %
0 %
100 %
0 %
100 %
0 %
100 %
0 %
Figure 18. Google Trend Visualization.
Furthermore, we link upcoming developments in the field of predictive maintenance to the
development of the Internet of Things. Three aspects are of particular interest in this context:
•
Algorithm Development. Current success stories on deep learning have not affected research yet.
In particular, neural networks were used in all periods. Please note that the introduction of
Appl. Sci. 2018,8, 918 22 of 29
convolutional neural networks (cnn) might improve predictions. So far, this technique has been
widely used for image classification [
91
], but the tendency of bringing more and more sensors
into industrial applications will make this multidimensional approach more attractive as well.
•
Trust. Using predictive systems, the importance of planning algorithms for industrial production is
increasing. Decisions are based on algorithm results and a evolution of these systems is leveraged
if people agree on calculations and if they trust their predictive systems. According to [
92
],
high agreement rates can be achieved by trust, which, in turn, depends on reliability. Following
the theory that improvements in algorithm result in a better reliability of predictive systems,
we expect better trust and acceptance rates as well.
•
Alliances. From Figure 14, it can be discovered that in Clusters 3 and 4, the countries working in
the same cluster are in a conflicting state. If deteriorating country relations influences research
progress, the pairs US-China, Russia-Germany, and Great Britain-Germany (Brexit) might
complicate research cooperations and consequently slow down the scientific progress due to
reduced knowledge exchange.
Altogether, we expect a further growth in the number of research publications in the field of
Predictive Maintenance.
7. Threats to Validity
The bibliometric analysis is a powerful tool. Though, some limitations need to be considered.
First, a general quality measure for choosing search queries is challenging. As the total number of
all published papers is unknown, it cannot be measured how good we cover the state of the art.
Furthermore, the comparison with other search query choices is difficult as we use preprocessing
filters. Our search strings result in many papers. Therefore, we improve the significance of centrality
measures, such as betweenness and degree. On the other hand, we take the risk to include irrelevant
papers in the analysis. Next, as there is an unlimited number of combinations for the parameters
of the bibliometric analysis, we cannot guarantee optimal cluster sizes. When the cluster size is too
small, related research fields are likely not to be representative for the dataset, i.e., they are too specific.
By contrast, by conducting the analysis with too large cluster sizes, we might merge two different
cluster labels and loose information as well as precision. Another thread to validity concerns the choice
of the number of citations as an indicator for the importance of research, since quality is not necessarily
expressed by public attention. Finally, the bibliometric analysis solely covers the academic part of
approaches. Knowledge that is present within enterprises, but which is not publicly available, is not
included in the analysis. Therefore, we cannot evaluate the feasibility of the presented approaches and
techniques.
8. Summary and Outlook
This work identified important papers and techniques used in the field of state of the art
equipment maintenance systems based on a bibliometric analysis. The latter covers temporal,
geographical, author-related, and technique-related issues. For this analysis, we used search strings
from contemporary maintenance methods to extract sets of papers from an academic database.
This work provides an overview on maintenance strategies on one hand and serves as a schema
to categorize future research results on the other by providing look-up tables for research fields and
techniques. The techniques, which were mostly identified by keywords, indicate how researchers
approach the topic of state of the art equipment maintenance systems. Furthermore, each research field is
represented by the most important paper, whose importance is determined by the number of citations.
The collection of lead documents extracted from different search strings is not comparable due to
different time periods and research fields.
We further introduced an importance measure to compare the papers by standardizing their
properties and by expressing their relevance in a more cluster-independent way. The weighting of
each parameter needed for calculating of the importance measure can be chosen by the conductor of
Appl. Sci. 2018,8, 918 23 of 29
the analysis. Therefore, this procedure is adaptive and can be reused in other bibliometric analyses.
Since state of the art equipment maintenance systems, mainly represented by predictive maintenance,
constitute a growing research field, it is important to stay informed about the current research.
This work allows for the classification of emerging works as well as for the comparison with existing
literature. As shown, the topic of equipment maintenance systems consists of many different research
fields and various applied techniques. For our future work, we will focus our research on modular
predictive maintenance applications. These applications, in turn, need to scale up with an increasing
number of sensors, as particularly required in the context of the Internet of Things. We will evaluate
the integration of the revealed techniques into new equipment maintenance systems modules. Finally,
further research will emphasize the modular design of predictive maintenance applications to support
scaling applications and flexible procedures.
Author Contributions:
B.H. initiated, conceived and wrote the paper. R.P. and R.M. corrected and supervised
this work. F.M.-B. and B.S. did the technical support for the bibliometric software. A.T. and K.K. offered resources
and drew conclusions.
Conflicts of Interest: The authors declare no conflict of interest.
Abbreviations
The following abbreviations are used in this manuscript:
AHP Analytic Hierarchy Process
FA Frequency Analysis
FEA Finite Element Analysis
IR Infrared
MCO Maintenance Cost Optimization
MPC Model Predictive Control
RCA Root Cause Analysis
SVM Support Vector Machine
Appendix A. Research Fronts
Appendix A.1
Table A1. Research field and technique of the most cited papers 2000–2008.
# Lead Document Citations Cluster Label Technique
1
Strategic alliances as social capital: a multidimensional
view [89]281 Predictive Models Interfirm Networks
2Mining Version Histories to Guide Software
Changes [93]196 Software Prediction ROSE framework
3 Intelligent prognostics tools and e-maintenance [94] 158 Monitoring e-Maintenance
4Continuous time predictive maintenance scheduling
for a deteriorating system [95]141 Cost Based Model Maintenance Cost
Optimization
5
The analytic hierarchy process applied to maintenance
strategy selection [39]133 Model Predictive Control Analytic Hierarchy
Process
6
Robust performance degradation assessment methods
for enhanced rolling element bearing prognostics [96]127 Vibration Analysis Self Organizing Map
7 Infrared thermography for building diagnostics [97] 119 Thermography IR Camera
8Applications of dielectric spectroscopy in time and
frequency domain for HV power equipment [98]108 Cable Diagnostic Frequency Analysis
9 Linking maintenance strategies to performance [74] 108 Managing PdM Total Productive
Maintenance
10 Root cause AC motor failure analysis [99] 44 Motor Fault Prediction Root Cause Analysis
11 Temperature development in structural stainless steel
sections exposed to fire [100]44 Generic Approach Heat Transfer
Coefficient
12 Finite element analysis of internal winding faults in
distribution transformers [101]35 Prediction on Electric
Systems
Finite Element
Analysis
13
Creating an abstraction of sensors to ease usage,
distribution and management of a measurement
network [102]
0 Sensor Data Analog Signature
Analysis
Appl. Sci. 2018,8, 918 24 of 29
Appendix A.2
To understand how clustering works, the computation of the second-order similarities will be
explained. First, all the terms of title, abstract, and author keywords are combined and preprocessed.
This procedure includes stemming (reducing words to their word stem), removing stopwords (like the,
a...), and filtering special characters. Let
fim
be the frequency of a term tin document
m
, while
N
represents the total number of documents and the number of documents containing tis denoted as
ji
.
The resulting weight of terms is called TF-IDF:
wtm =ftm ∗log N
nt!(A1)
By summing up the second-order weights and applying Salton’s Cosine for two documents
m
and p, with several terms k, we receive the first-order similarity:
smp =∑k
i=1wim ·wip
q∑k
i=1w2
im ·q∑k
i=1w2
ip
(A2)
To have a connection among all documents, we execute the cosine index once more and
obtain the second-order similarity matrix
s2,mp
, which is filtered in a final step with the k-nearest-
neighbour method.
s2,mp =∑N
j=1sjm ·sjp
q∑N
j=1s2
jm ·q∑N
j=1s2
jp
(A3)
Appendix A.3
Table A2. Research field and technique of the most cited papers 2008–2016.
# Article Cited Cluster Label Technique
1A survey of the application of gamma processes in
maintenance [103]241 Multi-component Markov Processes
2 Classifying Software Changes: Clean or Buggy? [104] 103 Software prediction Support Vector
Machine
3The state of the art in chemical process control in
Japan: Good practice and questionnaire survey [105]87 Model-predictive Control PID Control
4The prediction and diagnosis of wind turbine
faults [106]84 Wind Turbines Data-Mining
5Current status of machine prognostics in
condition-based maintenance: a review [107]71 Prognostics and Health
Management Decision Framework
6
The differing and mediating roles of trust and
relationship commitment in service relationship
maintenance and development [108]
58 Cloud and IoT Covariance Structure
Model
7
Rolling element bearing fault diagnosis based on the
combination of genetic algorithms and fast
kurtogram [109]
58 Rolling Bearing Genetic algorithm
8
A Real-Time Predictive-Maintenance System of
Aluminum Electrolytic Capacitors Used in
Uninterrupted Power Supplies [110]
35 Capacitor Kalman Filtering
9
Intelligent state estimation for fault tolerant nonlinear
predictive control [111]28 Thermography Kalman Filter
10 Machine and component residual life estimation
through the application of neural networks [112]19 Generic approach Neural Network
11
Concrete cover cracking caused by steel reinforcement
corrosion [113]18 Corrosion Corrosion Rate
12 An Advanced Stator Winding Insulation Quality
Assessment Technique for Inverter-Fed Machines [
114
]
17 Vibration Analysis Dissipation Factor
13 On-line monitoring applications in nuclear power
plants [73]17 Nuclear Risks and Safety Noise Analysis
14 Development of improved railway track degradation
models [115]15 Rail Maintenance Degradation Model
15 Aged oils reclamation: Facts and arguments based on
laboratory studies [116]7 Oils Fuller’s Earth
Appl. Sci. 2018,8, 918 25 of 29
References
1.
Saxena, A.; Goebel, K.; Simon, D.; Eklund, N. Damage propagation modeling for aircraft engine run-to-failure
simulation. In Proceedings of the International Conference on Prognostics and Health Management, Denver,
CO, USA, 6–9 October 2018.
2. Barlow, R.; Hunter, L. Optimum preventive maintenance policies. Oper. Res. 1960,8, 90–100. [CrossRef]
3. Mobley, R.K. An Introduction to Predictive Maintenance; Elsevier Science: New York City, NY, USA, 2002.
4.
Grant, M.J.; Booth, A. A typology of reviews: An analysis of 14 review types and associated methodologies.
Health Inf. Libr. J. 2009,26, 91–108. [CrossRef] [PubMed]
5.
Muller, A.; Crespo Marquez, A.; Iung, B. On the concept of e-maintenance: Review and current research.
Reliab. Eng. Syst. Saf. 2008,93, 1165–1187. [CrossRef]
6.
Wang, H. A survey of maintenance policies of deteriorating systems. Eur. J. Oper. Res.
2002
,139, 469–489.
[CrossRef]
7.
Pierskalla, W.P.; Voelker, J.A. A survey of maintenance models: The control and surveillance of deteriorating
systems. Naval Res. Logist. Q. 1976,23, 353–388. [CrossRef]
8.
Cho, D.I.; Parlar, M. A survey of maintenance models for multi-unit systems. Eur. J. Oper. Res.
1991
,51, 1–23.
[CrossRef]
9.
Lu, B.; Li, Y.; Wu, X.; Yang, Z. A review of recent advances in wind turbine condition monitoring and fault
diagnosis. In Proceedings of the IEEE Power Electronics and Machines in Wind Applications, Lincoln, NE,
USA, 24–26 June 2009; pp. 1–7.
10.
Barnish, T.; Muller, M.; Kasten, D. Motor Maintenance: A Survey of Techniques and Results; American Council
for an Energy-Efficient Economy: Washington, DC, USA, 1997.
11.
Garg, A.; Deshmukh, S.G. Maintenance management: Literature review and directions. J. Qual. Maint. Eng.
2006,12, 205–238. [CrossRef]
12.
Chadegani, A.A.; Salehi, H.; Yunus, M.M.; Farhadi, H.; Fooladi, M.; Farhadi, M.; Ebrahim, N.A.
A Comparison between Two Main Academic Literature Collections: Web of Science and Scopus Databases.
Asian Soc. Sci. 2013,9.10.5539/ass.v9n5p18. [CrossRef]
13.
Amadi-Echendu, J.E.; Willett, R.; Brown, K.; Hope, T.; Lee, J.; Mathew, J.; Vyas, N.; Yang, B.S. What is
engineering asset management? In Definitions, Concepts and Scope of Engineering Asset Management; Springer:
Berlin, Germany, 2010; pp. 3–16.
14.
Koronios, A.; Nastasie, D.; Chanana, V.; Haider, A. Integration through standards—An overview of
international standards for engineering asset management. In Proceedings of the Fourth International
Conference on Condition Monitoring, Harrogate, UK, 11–14 June 2007.
15.
Stapelberg, R. Professional skills training in integrated asset management: how to develop and implement
the essential organisational asset management functions. In Engineering Asset Management; Springer: Berlin,
Germany, 2006; pp. 1243–1251.
16.
Ma, L. Condition monitoring in engineering asset management. In Proceedings of the Asia Pacific Vibration
Conference, Sapporo, Japan, 6–9 August 2007.
17. Pritchard, A. Statistical Bibliography or Bibliometrics? J. Doc. 1969,25, 348–349.
18. de Solla Price, D.J. Networks of Scientific Papers. Science 1965,149, 510–515. [CrossRef]
19.
Schiebel, E. Visualization of research fronts and knowledge bases by three-dimensional areal densities of
bibliographically coupled publications and co-citations. Scientometrics 2012,91, 557–566. [CrossRef]
20.
Liu, X.; Bollen, J.; Nelson, M.L.; van de Sompel, H. Co-authorship networks in the digital library research
community. Inf. Process. Manag. 2005,41, 1462–1480. [CrossRef]
21.
Stelzer, B.; Meyer-Brötz, F.; Schiebel, E.; Brecht, L. Combining the scenario technique with bibliometrics for
technology foresight: The case of personalized medicine. Technol. Forecast. Soc. Chang.
2015
,98, 137–156.
[CrossRef]
22.
Cobo, M.J.; López-Herrera, A.G.; Herrera-Viedma, E.; Herrera, F. Science mapping software tools: Review,
analysis, and cooperative study among tools. J. Assoc. Inf. Sci. Technol. 2011,62, 1382–1402. [CrossRef]
23.
Meyer-Brötz, F.; Stelzer, B.; Schiebel, E.; Brecht, L. Mapping the Technology and Innovation Management Literature
Using Hybrid Bibliometric Networks; Inderscience: Olney, UK, 2017.
Appl. Sci. 2018,8, 918 26 of 29
24.
Meyer-Brötz, F.; Schiebel, E.; Brecht, L. Experimental evaluation of parameter settings in calculation of hybrid
similarities: effects of first- and second-order similarity, edge cutting, and weighting factors. Scientometrics
2017,111, 1307–1325. [CrossRef]
25.
Blondel, V.D.; Guillaume, J.L.; Lambiotte, R.; Lefebvre, E. Fast unfolding of communities in large networks.
J. Stat. Mech. 2008,2008, P10008. [CrossRef]
26.
Martin, S.; Brown, W.M.; Klavans, R.; Boyack, K.W. OpenOrd: An Open-Source Toolbox for Large Graph
Layout. In Proceedings of the SPIE 7868 Visualization and Data Analysis, San Francisco Airport, CA, USA,
23–27 January 2011.
27.
Fruchterman, T.M.J.; Reingold, E.M. Graph drawing by force-directed placement. Software
1991
,21, 1129–1164.
[CrossRef]
28.
Meyer-Brötz, F. Bibliometric Timelines for the Identification of Emerging Technologies; Springer: Berlin,
Gremany, 2017.
29.
Woon, W.L.; Zeineldin, H.; Madnick, S. Bibliometric Analysis of Distributed Generation. Technol. Forecast.
Soc. Chang. 2009,78, 408–420. [CrossRef]
30.
Davies, A. Handbook of Condition Monitoring: Techniques and Methodology; Springer Netherlands: Dordrecht,
The Netherlands, 2012.
31.
Pantsar-Syvaniemi, S.; Ovaska, E.; Ferrari, S.; Cinotti, T.S.; Zamagni, G.; Roffia, L.; Mattarozzi, S.; Nannini, V.
Case Study: Context-Aware Supervision of a Smart Maintenance Process. In Proceedings of the IEEE/IPSJ
11th International Symposium on Applications and the Internet (SAINT), Munich, Germany, 18–21 July 2011;
pp. 309–314.
32.
Smith, R.; Hawkins, B. Lean Maintenance: Reduce Costs, Improve Quality, and Increase Market Share;
Butterworth-Heinemann: Oxford, UK, 2004.
33. Moubray, J. Reliability-Centered Maintenance, 2nd ed.; Industrial Press: New York, NY, USA, 1997.
34.
Siegel, E. Predictive analytics: The Power to Predict Who Will Click, Buy, Lie, or Die; John Wiley & Sons
Incorporated: Hoboken, NY, USA, 2016.
35. Trotter, J.A. Techniques of Preictive Maintenance. Chem. Eng. 1970,77, 66.
36.
Hirsch, J.E. An index to quantify an individual’s scientific research output. Proc Natl. Acad. Sci. USA
2005
,
102, 16569–16572. [CrossRef] [PubMed]
37.
Mehta, S.V.; Uzsoy, R.M. Predictable scheduling of a job shop subject to breakdowns. IEEE Trans. Robot.
Autom. 1998,14, 365–378. [CrossRef]
38.
Fyfe, K.R.; Munck, E. Analysis of computed order tracking. Mech. Syst. Signal Process.
1997
,11, 187–205.
[CrossRef]
39.
Bevilacqua, M.; Braglia, M. The analytic hierarchy process applied to maintenance strategy selection. Reliab.
Eng. Syst. Saf. 2000,70, 71–83. [CrossRef]
40.
Cash, M.A.; Habetler, T.G.; Kliman, G.B. Insulation failure prediction in AC machines using line-neutral
voltages. IEEE Trans. Ind. Appl. 1998,34, 1234–1239. [CrossRef]
41.
Golabi, K.; Shepard, R. Pontis: A System for Maintenance Optimization and Improvement of US Bridge
Networks. Interfaces 1997,27, 71–88. [CrossRef]
42.
Ebert, C. Classification techniques for metric-based software development. Softw. Qual. J.
1996
,5, 255–272.
[CrossRef]
43.
Gelormino, M.S.; Ricker, N.L. Model-predictive control of a combined sewer system. Int. J. Control
1994
,
59, 793–816. [CrossRef]
44.
Madding, R.P. Emissivity measurement and temperature correction accuracy considerations. In Proceedings
of the International Society for Optics and Photonics, Orlando, FL, USA, 5–9 April 1999; pp. 393–401.
45.
Lee, S.F.; Spanos, C.J. Prediction of wafer state after plasma processing using real-time tool data. IEEE Trans.
Semicond. Manuf. 1995,8, 252–261. [CrossRef]
46. Kanter, R.M. Collaborative Advantage—The art of alliances. Harv. Bus. Rev. 1994,72, 96–108.
47. Rausand, M. Reliability centered maintenance. Reliab. Eng. Syst. Saf. 1998,60, 121–132. [CrossRef]
48.
Harunuzzaman, M.; Aldemir, T. Optimization of standby safety system maintenance schedules in nuclear
power plants. Nucl. Technol. 1996,113, 354–367. [CrossRef]
49.
Yang, J.N.; Manning, S.D. Stochastic crack growth analysis methodologies for metallic structures. Eng. Fract.
Mech. 1990,37, 1105–1124. [CrossRef]
Appl. Sci. 2018,8, 918 27 of 29
50.
Shah, R.; Ward, P.T. Lean manufacturing: context, practice bundles, and performance. J. Oper. Manag.
2003
,
21, 129–149. [CrossRef]
51.
Endrenyi, J.; Aboresheid, S.; Allan, R.N.; Anders, G.J.; Asgarpoor, S.; Billinton, R.; Chowdhury, N.;
Dialynas, E.N.; Fipper, M.; Fletcher, R.H.; et al. The present status of maintenance strategies and the
impact of maintenance on reliability. IEEE Trans. Power Syst. 2001,16, 638–646. [CrossRef]
52.
Kiritsis, D.; Bufardi, A.; Xirouchakis, P. Research issues on product lifecycle management and information
tracking using smart embedded systems. Intell. Maint. Syst. 2003,17, 189–202. [CrossRef]
53.
Kothamasu, R.; Huang, S.H.; VerDuin, W.H. System health monitoring and prognostics—A review of current
paradigms and practices. Int. J. Adv. Manuf. Technol. 2006,28, 1012–1024. [CrossRef]
54.
Zhou, X.; Xi, L.; Lee, J. Reliability-centered predictive maintenance scheduling for a continuously monitored
system subject to degradation. Reliab. Eng. Syst. Saf. 2007,92, 530–534. [CrossRef]
55.
Avdelidis, N.P.; Almond, D.P.; Dobbinson, A.; Hawtin, B.C.; Ibarra-Castanedo, C.; Maldague, X. Aircraft
composites assessment by means of transient thermal NDT. Prog. Aerosp. Sci.
2004
,40, 143–162. [CrossRef]
56.
Ramamurthy, H.; Prabhu, B.S.; Gadh, R.; Madni, A.M. Wireless Industrial Monitoring and Control Using a
Smart Sensor Platform. IEEE Sens. J. 2007,7, 611–618. [CrossRef]
57.
Depuru, S.S.; Wang, L.; Devabhaktuni, V. Smart meters for power grid: Challenges, issues, advantages and
status. Renew. Sustain. Energy Rev. 2011,15, 2736–2742. [CrossRef]
58.
Lee, J.; Wu, F.; Zhao, W.; Ghaffari, M.; Liao, L.; Siegel, D. Prognostics and health management design
for rotary machinery systems-Reviews, methodology and applications. Mech. Syst. Signal Process.
2014
,
42, 314–334. [CrossRef]
59.
Shmueli, G.; Koppius, O.R. Predictive analytics in information systems research. Mis Q.
2011
,35, 553–572.
[CrossRef]
60.
Candido, G.; Colombo, A.W.; Barata, J.; Jammes, F. Service-Oriented Infrastructure to Support the
Deployment of Evolvable Production Systems. IEEE Trans. Ind. Inform. 2011,7, 759–767. [CrossRef]
61.
Waller, M.A.; Fawcett, S.E. Data Science, Predictive Analytics, and Big Data: A Revolution That Will
Transform Supply Chain Design and Management. J. Bus. Logist. 2013,34, 77–84. [CrossRef]
62.
Niu, G.; Yang, B.S.; Pecht, M. Development of an optimized condition-based maintenance system by data
fusion and reliability-centered maintenance. Reliab. Eng. Syst. Saf. 2010,95, 786–796. [CrossRef]
63.
Shah, R.; Chandrasekaran, A.; Linderman, K. In pursuit of implementation patterns: The context of Lean
and Six Sigma. Int. J. Prod. Res. 2008,46, 6679–6699. [CrossRef]
64.
Fischer, K.; Besnard, F.; Bertling, L. Reliability-Centered Maintenance for Wind Turbines Based on Statistical
Analysis and Practical Experience. IEEE Trans. Energy Convers. 2012,27, 184–195. [CrossRef]
65.
Li, V.C.; Herbert, E. Robust Self-Healing Concrete for Sustainable Infrastructure. J. Adv. Concr. Technol.
2012
,
10, 207–218. [CrossRef]
66.
Russell, B.D.; Benner, C.L. Intelligent Systems for Improved Reliability and Failure Diagnosis in Distribution
Systems. IEEE Trans. Smart Grid 2010,1, 48–56. [CrossRef]
67.
Ahmad, R.W.; Gani, A.; Hamid, S.H.A.; Shiraz, M.; Yousafzai, A.; Xia, F. A survey on virtual machine
migration and server consolidation frameworks for cloud data centers. J. Netw. Comput. Appl.
2015
,52, 11–25.
[CrossRef]
68.
Liu, Z.; Kleiner, Y. State of the art review of inspection technologies for condition assessment of water pipes.
Measurement 2013,46, 1–15. [CrossRef]
69.
Ingram, D.M.E.; Steinhauser, F.; Marinescu, C.; Taylor, R.R.; Schaub, P.; Campbell, D.A. Direct Evaluation of
IEC 61850-9-2 Process Bus Network Performance. IEEE Trans. Smart Grid 2012,3, 1853–1854. [CrossRef]
70. Zakeri, B.; Syri, S. Electrical energy storage systems: A comparative life cycle cost analysis. Renew. Sustain.
Energy Rev. 2015,42, 569–596. [CrossRef]
71.
Stoneburner, G.; Goguen, A.; Feringa, A. Risk Management Guide for Information Technology Systems.
Inf. Syst. Secur. Risk Model 2002,800, 800–830.
72.
Si, X.S.; Wang, W.; Hu, C.H.; Zhou, D.H. Remaining useful life estimation—A review on the statistical data
driven approaches. Eur. J. Oper. Res. 2011,213, 1–14. [CrossRef]
73.
Hashemian, H.M. On-line monitoring applications in nuclear power plants. Prog. Nucl. Energy
2011
,
53, 167–181. [CrossRef]
74.
Swanson, L. Linking maintenance strategies to performance. Int. J. Prod. Econ.
2001
,70, 237–244. [CrossRef]
Appl. Sci. 2018,8, 918 28 of 29
75.
Nandi, S.; Toliyat, H.A.; Li, X.D. Condition monitoring and fault diagnosis of electrical motors—A review.
IEEE Trans. Energy Convers. 2005,20, 719–729. [CrossRef]
76.
Jardine, A.K.S.; Lin, D.; Banjevic, D. A review on machinery diagnostics and prognostics implementing
condition-based maintenance. Mech. Syst. Signal Process. 2006,20, 1483–1510. [CrossRef]
77.
Widodo, A.; Yang, B.S. Support vector machine in machine condition monitoring and fault diagnosis. Mech.
Syst. Signal Process. 2007,21, 2560–2574. [CrossRef]
78.
Yu, J.; Qin, S.J. Multimode process monitoring with Bayesian inference-based finite Gaussian mixture models.
AICHE J. 2008,54, 1811–1829. [CrossRef]
79.
Hameed, Z.; Hong, Y.S.; Cho, Y.M.; Ahn, S.H.; Song, C.K. Condition monitoring and fault detection of wind
turbines and related algorithms: A review. Renew. Sustain. Energy Rev. 2009,13, 1–39. [CrossRef]
80.
Pantelopoulos, A.; Bourbakis, N.G. A Survey on Wearable Sensor-Based Systems for Health Monitoring and
Prognosis. IEEE Trans. Syst. Man Cybern. Part C 2010,40, 1–12. [CrossRef]
81.
Yin, S.; Ding, S.X.; Haghani, A.; Hao, H.; Zhang, P. A comparison study of basic data-driven fault diagnosis
and process monitoring methods on the benchmark Tennessee Eastman process. J. Process Control
2012
,
22, 1567–1581. [CrossRef]
82.
Bagavathiappan, S.; Lahiri, B.B.; Saravanan, T.; Philip, J.; Jayakumar, T. Infrared thermography for condition
monitoring—A review. Infrared Phys. Technol. 2013,60, 35–55. [CrossRef]
83.
Yin, S.; Ding, S.X.; Xie, X.; Luo, H. A Review on Basic Data-Driven Approaches for Industrial Process
Monitoring. IEEE Trans. Ind. Electron. 2014,61, 6418–6428. [CrossRef]
84.
Yin, S.; Li, X.; Gao, H.; Kaynak, O. Data-Based Techniques Focused on Modern Industry: An Overview. IEEE
Trans. Ind. Electron. 2015,62, 657–667. [CrossRef]
85.
Zhang, X.; Clark, M.; Rattan, K.; Muse, J.; Khalili, M. Controller Integrity Monitoring in Adaptive Learning
Systems Towards Trusted Autonomy. IEEE Trans. Autom. Sci. Eng. 2016,13, 491–501. [CrossRef]
86.
Bollen, J.; van de Sompel, H.; Hagberg, A.; Chute, R. A principal component analysis of 39 scientific impact
measures. PLoS ONE 2009,4, e6022. [CrossRef] [PubMed]
87.
Hakimi, S.L. On Realizability of a Set of Integers as Degrees of the Vertices of a Linear Graph. I. J. Soc. Ind.
Appl. Math. 1962,10, 496–506. [CrossRef]
88. Brandes, U. A Faster Algorithm for Betweenness Centrality. J. Math. Sociol. 2001,25, 163–177. [CrossRef]
89.
Koka, B.R.; Prescott, J.E. Strategic alliances as social capital: A multidimensional view. Strateg. Manag. J.
2002,23, 795–816. [CrossRef]
90.
Bishop, P.; Hines, A.; Collins, T. The current state of scenario development: An overview of techniques.
Foresight 2007,9, 5–25. [CrossRef]
91.
Ciresan, D.; Meier, U.; Schmidhuber, J. Multi-column deep neural networks for image classification.
In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Providence, RI,
USA, 16–21 June 2012; pp. 3642–3649.
92.
Chancey, E.T.; Bliss, J.P.; Proaps, A.B.; Madhavan, P. The Role of Trust as a Mediator Between System
Characteristics and Response Behaviors. Hum. Factors 2015,57, 947–958. [CrossRef] [PubMed]
93.
Zimmermann, T.; Weisgerber, P.; Diehl, S.; Zeller, A. Mining Version Histories to Guide Software Changes.
IEEE Trans. Softw. Eng. 2004,31, 429–445. [CrossRef]
94.
Lee, J.; Ni, J.; Djurdjanovic, D.; Qiu, H.; Liao, H. Intelligent prognostics tools and e-maintenance. Comput.
Ind. 2006,57, 476–489. [CrossRef]
95.
Dieulle, L.; Berenguer, C.; Grall, A.; Roussignol, M. Continuous time predictive maintenance scheduling
for a deteriorating system. In Proceedings of the Annual Reliability and Maintainability Symposium,
the International Symposium on Product Quality & Integrity, Philadelphia, PA, USA, 22–25 January 2001;
IEEE Operations Center: Piscataway, NJ, USA, 2001; pp. 150–155.
96.
Qiu, H.; Lee, J.; Lin, J.; Yu, G. Robust performance degradation assessment methods for enhanced rolling
element bearing prognostics. Intell. Maint. Syst. 2003,17, 127–140. [CrossRef]
97.
Balaras, C.A.; Argiriou, A.A. Infrared thermography for building diagnostics. Energy Build.
2002
,
34, 171–183. [CrossRef]
98.
Zaengl, W.S. Applications of dielectric spectroscopy in time and frequency domain for HV power equipment.
IEEE Electr. Insul. Mag. 2003,19, 9–22. [CrossRef]
Appl. Sci. 2018,8, 918 29 of 29
99.
Bonnett, A.H. Root cause AC motor failure analysis. In Proceedings of the Industry Applications Society
46th Annual Petroleum and Chemical Industry Conference, San Diego, CA, USA, 13–15 September 1999;
pp. 85–97.
100.
Gardner, L.; Ng, K.T. Temperature development in structural stainless steel sections exposed to fire. Fire Saf. J.
2006,41, 185–203. [CrossRef]
101.
Wang, H.; Butler, K.L. Finite element analysis of internal winding faults in distribution transformers. IEEE
Trans. Power Deliv. 2001,16, 422–428. [CrossRef]
102.
Marino, P.; Siguenza, C.; Poza, F.; Vazquez, F.; Machado, F. Creating an abstraction of sensors to ease usage,
distribution and management of a measurement network. In Proceedings of the 2003 IEEE International
Conference on Emerging Technologies and Factory Automation, Lisbon, Portugal, 16–19 September 2003;
pp. 471–478.
103.
van Noortwijk, J.M. A survey of the application of gamma processes in maintenance. Maint. Model. Appl.
2009,94, 2–21. [CrossRef]
104. Kim, S.; Whitehead, E.J.; Zhang, Y. Classifying Software Changes: Clean or Buggy? IEEE Trans. Softw. Eng.
2008,34, 181–196.
105.
Kano, M.; Ogawa, M. The state of the art in chemical process control in Japan: Good practice and
questionnaire survey. J. Process Control 2010,20, 969–982. [CrossRef]
106.
Kusiak, A.; Li, W. The prediction and diagnosis of wind turbine faults. Renew. Energy
2011
,36, 16–23.
[CrossRef]
107.
Peng, Y.; Dong, M.; Zuo, M.J. Current status of machine prognostics in condition-based maintenance:
A review. Int. J. Adv. Manuf. Technol. 2010,50, 297–313. [CrossRef]
108.
Aurier, P.; N’Goala, G. The differing and mediating roles of trust and relationship commitment in service
relationship maintenance and development. J. Acad. Mark. Sci. 2010,38, 303–325. [CrossRef]
109.
Zhang, Y.; Randall, R.B. Rolling element bearing fault diagnosis based on the combination of genetic
algorithms and fast kurtogram. Mech. Syst. Signal Process. 2009,23, 1509–1517. [CrossRef]
110.
Abdennadher, K.; Venet, P.; Rojat, G.; Retif, J.M.; Rosset, C. A Real-Time Predictive-Maintenance System
of Aluminum Electrolytic Capacitors Used in Uninterrupted Power Supplies. IEEE Trans. Ind. Appl.
2010
,
46, 1644–1652. [CrossRef]
111.
Deshpande, A.P.; Patwardhan, S.C.; Narasimhan, S.S. Intelligent state estimation for fault tolerant nonlinear
predictive control. J. Process Control 2009,19, 187–204. [CrossRef]
112.
Herzog, M.A.; Marwala, T.; Heyns, P.S. Machine and component residual life estimation through the
application of neural networks. Reliab. Eng. Syst. Saf. 2009,94, 479–489. [CrossRef]
113.
Al-Harthy, A.S.; Stewart, M.G.; Mullard, J. Concrete cover cracking caused by steel reinforcement corrosion.
Mag. Concr. Res. 2011,63, 655–667. [CrossRef]
114.
Yang, J.; Cho, J.; Lee, S.B.; Yoo, J.Y.; Kim, H.D. An Advanced Stator Winding Insulation Quality Assessment
Technique for Inverter-Fed Machines. IEEE Trans. Ind. Appl. 2008,44, 555–564. [CrossRef]
115.
Sadeghi, J.; Askarinejad, H. Development of improved railway track degradation models. Struct. Infrastruct.
Eng. 2010,6, 675–688. [CrossRef]
116.
N’cho, J.S.; Fofana, I.; Beroual, A.; Aka-Ngnui, T.; Sabau, J. Aged oils reclamation: Facts and arguments
based on laboratory studies. IEEE Trans. Dielectr. Electr. Insul. 2012,19, 1583–1592. [CrossRef]
c
2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
