Available online at www.sciencedirect.

com

ScienceDirect
Procedia CIRP 72 (2018) 1039–1044
www.elsevier.com/locate/procedia

51st CIRP Conference on Manufacturing Systems

PriMa-X: A reference model for realizing prescriptive maintenance and

assessing its maturity enhanced by machine learning
Tanja Nemetha,b,*, Fazel Ansari a,b, Wilfried Sihn a,b, Bernhard Haslhoferc, Alexander Schindlerc

a
Vienna University of Technology, Institute of Management Science, Vienna 1040, Austria
b
Fraunhofer Austria Research GmbH, Division of Production and Logistics Management, Vienna 1040, Austria
c
Austrian Institute of Technology, Digital Insight Lab, Vienna, Austria

* Corresponding author. Tel.: +43-676-888-616-19; Fax: +43-1-58801-330-99. E-mail address: [email protected]

Abstract

The digital transformation already has a strong impact on manufacturing techniques and processes and requires novel data-driven maintenance
strategies and models, which support prompt and effective decision-making. This poses new requirements, challenges and opportunities for
securing and improving machine availability and process stability. This paper builds on the concept of prescriptive maintenance and proposes a
reference model that (i) supports the implementation of a prescriptive maintenance strategy and the assessment of its maturity level, (ii) facilitates
the integration of data-science methods for predicting future events, and (iii) identifies action fields to reach an enhanced target maturity state
and thus higher prediction accuracy.

© 2018 The Authors. Published by Elsevier B.V.

Peer-review under responsibility of the scientific committee of the 51st CIRP Conference on Manufacturing Systems.

Keywords: cyber physical production systems; prescriptive maintenance; data science; reference model; maturity

1. Introduction Hence, a paradigm shift from descriptive to predictive and

prescriptive maintenance is triggered [10]. The concept of
The industrialized world is currently facing a fourth prescriptive maintenance extends beyond the mere prediction
revolution through the realization of digitalized manufacturing, of failures. Utilizing sensing technology, modeling expert
which is built upon cutting-edge computer science, information knowledge, and employing predictive data analytics, based on
and communication technologies as well as manufacturing historical and incoming real time data, allow making
science and technology [1]. These advances have paved the way predictions on when a failure occurs and ultimately
for systematical deployment of Cyber Physical Production recommending an optimal course of action [11, 10]. According
Systems (CPPS) [2]. Utilizing CPPS significantly contributes to Ansari et. al. (2017) prescriptive maintenance of CPPS not
in increasing productivity, machine availability and automation only aims at understanding and reasoning out past events, but
level, improving resource efficiency and ensuring product also at anticipating the likelihood of future events and potential
quality of a manufacturing system [1, 3]. Remote and real-time effects of each decision alternative on the physical space and
control, complete embedded systems, predictability and associated business processes [10]. The two most important
robustness at every level as well as safety are further complementary skills required for implementing prescriptive
expectations associated to CPPS [4, 5, 6]. As a result, increasing maintenance in practice are: production process and system
complexity in terms of products, processes and machines [7, 8, knowledge on the one hand and data science methods and skills
9] arise, which dramatically increase the need for rethinking on the other hand [6].
and reshaping maintenance management organizations, models Although several studies emphasize the importance of
and associated systems [10]. predictive and prescriptive maintenance to overcome current

2212-8271 © 2018 The Authors. Published by Elsevier B.V.

Peer-review under responsibility of the scientific committee of the 51st CIRP Conference on Manufacturing Systems.
10.1016/j.procir.2018.03.280

This is a resupply of March 2023 as the template used in the publication of the original article contained errors. The content of the article has remained unaffected.
 1040 Tanja Nemeth et al. / Procedia CIRP 72 (2018) 1039–1044

challenges of digitalized manufacturing [12], hardly any of  Diagnostic maintenance answers the question “Why did it
these strategies have been implemented in practice. Recent happen?” by analyzing cause-effect relations, reasoning,
empirical studies revealed that 15% of producing companies and providing further technical details about former
pursue partly predictive maintenance strategies, and maintenance operations.
prescriptive approaches are only applied by 4% [13]. Due to  Predictive maintenance answers the question “What will
vague (not precisely determined or distinguished) or ambiguous happen when?” by learning from historical maintenance
(have two or more interpretations) definitions and descriptions data, possibly in real-time, and predicting future events.
of these terms, companies fail in pursuing innovative This is also referred to as “Smart Maintenance”, “Data-
maintenance strategies [10]. Driven Maintenance” and recently as “Maintenance 4.0”.
In order to resolve the aforementioned challenges, a  Prescriptive maintenance answers the question “How can
prescriptive maintenance reference model is proposed within we make it happen?” or in other words “How can we
this paper, trying to answer the following research questions: i) control the occurrence of a specific event?” by providing
Which tasks and objectives should an efficient prescriptive actionable recommendations for decision making and
maintenance (operational and management) process comprise? improving and/or optimizing forthcoming maintenance
ii) Which data-science methods are most suitable for predicting processes. It also refers to the recent advances in
future events and how can they be integrated into a prescriptive enhancing self-organization capabilities of CPPS, which
maintenance process? and ultimately iii) How can the ideally aim at machine self-diagnosis and scheduled
prescriptive maintenance maturity level be assessed including maintenance.
operational and management parameters?
2.2. Maintenance Decision Support Models
2. State of the Art
The state-of-the-art literature review reveals that the
2.1. Knowledge-Based Maintenance Strategies majority of existing maintenance models aims at supporting
decision-making processes. By combining different data-
Prescriptive maintenance is known as the highest maturity sources and knowledge assets and applying data-science
and complexity level of knowledge-based maintenance (KBM) methods such as exploratory data analysis or machine learning,
[10]. KBM assumes that competitive advantages for stabilizing maintenance system intelligence is improved. A selection of
maintenance processes and reducing unplanned costs are recently developed maintenance decision support models
achieved through holistic consideration of production (MDSM) is presented in this section.
processes, rather than atomistic inspection of (all) influential Glawar et. al. (2016) outlined a holistic and anticipatory
components [10, 14]. Thus, KBM concentrates on analyzing framework, which “enables the identification of maintenance-
maintenance as a non-isolated sub-domain of production critical conditions and the prediction of failure moments and
systems, which, in turn, influences the organizational value quality deviations” of tooling machines [15]. Aghezzaf et. al
creation [14]. Recent investigations show that especially the (2016) addressed a degradation-based selective maintenance
sub-domains quality management, maintenance and production decision problem of a continuously monitored multi-
planning strongly interact and jointly determine the component system. By modelling components as time-
achievement of the desired production performance, equipment dependent stochastic processes a cost-optimal set of necessary
availability, and product quality [8, 9]. maintenance actions is found [16]. Moreover, Wang et. al
(2017) investigated “a cloud-based paradigm of predictive
maintenance based on mobile agent to enable timely
information acquisition, sharing and utilization for improved
accuracy and reliability in fault diagnosis, remaining service
life prediction, and maintenance scheduling.” [17]. Arab et. al
(2013) solved a dynamic maintenance scheduling problem for
a multi-component production system by taking into account
real-time information from workstations including remaining
reliability of equipment, cycle times, buffers’ capacity and
mean time to repair of machines [18].
Fig. 1. Knowledge-Based Maintenance Strategies [10] Besides, Bärenfänger-Wojciechowski et. al. (2017)
presented a reference integrated management approach, named
The central objective of KBM is to develop a generic smart maintenance, which combines key maintenance
concept for optimizing maintenance processes through knowledge assets, namely humans, sensors, data-management
comprehensive consideration of maintenance consequences, and assistance-systems [19]. The concept “knowledge as a
system conditions, organization, and processes [14]. Existing service”, was introduced by Abramovici et. al. (2017) and
approaches for achieving the aforementioned goals of KBM supports knowledge allocation and the recommendation of
can be categorized as follows (cf. Fig. 1) [10]: possible solutions in accordance with failure causes and
 Descriptive maintenance answers the question “What similarity degree between former failure descriptions stored in
happened?” by providing information about previous a semantic knowledge base [20]. Last but not least,
maintenance operations. A. K. Muchiri et. al (2017) developed a theoretical framework
for evaluating the efficiency of maintenance actions based on a

This is a resupply of March 2023 as the template used in the publication of the original article contained errors. The content of the article has remained unaffected.
 Tanja Nemeth et al. / Procedia CIRP 72 (2018) 1039–1044 1041

technical, managerial and human perspective [21], Mehairjan Active learning [30] is a possible technique for training
et. al (2016) developed a maintenance management maturity machine learning algorithms in settings with low availability of
model based on five holistic dimensions including data quality labelled training data (e.g., machine outages). The key idea is
[22] and Schumacher et. al (2016) developed an Industry 4.0 that an algorithm is allowed to choose data from which it learns.
maturity model, which indirectly assesses aspects relevant for However, to the best of our knowledge, the application of active
data-driven maintenance [23]. learning technique has not yet been investigated for real-world
Although these models achieve valuable results in the area prescriptive maintenance problems.
of prescriptive maintenance, they suffer from the following
shortcomings: i) they consider dynamics of maintenance 2.4. Literature Synthesis: Research Gap on Maturity
processes (using time variables) but do not fully or partially Assessment
consider learning and prediction of the behavior of process Considering the literature review presented in section 2.1-
related parameters overtime, ii) they are use-case-specific 2.3, we identified four dimensions, namely, KBM, MDSM,
(unique problem) and hardly generalizable to similar groups of machine learning (ML) and maturity assessment (MA), which
problems, iii) they do not adequately include feedback loops constitute the basis for prescriptive maintenance of CPPS (cf.
and efficiency assessment techniques in order to improve Tab. 1). There are several papers, which either cover the three
maintenance planning quality, and iv) they use well- dimensions of KBM, MDSM and ML or discuss the idea of
established, but outdated process models for data analysis and (maintenance) maturity assessment. Despite advantages, we
knowledge discovery, which clearly need to be refined and have determined a gap on integrated data-driven maintenance
extended for predictive analytics tasks. maturity assessment considering multi-dimensionality of
2.3. Data Science Methods for Prescriptive Maintenance maintenance problems affected by strategic, tactical and
operational processes. To the best of our knowledge, a
Groundwork for prescriptive maintenance data science systematic maintenance maturity model, which assesses the
methods has been laid by the field of condition monitoring, level of maturity from the angle of industrial data science,
which provides condition indicators and analysis methods for rather than considering only maintenance influencing factors
regular monitoring of actual mechanical conditions, operating such as organizational, cultural, IT and infrastructural factors
efficiency, and other indicators of machines and process (generic approaches), is nonexistent.
systems. These conditions feed into a wide spectrum of possible
Table 1. Literature Synthesis
maintenance techniques, which include vibration monitoring,
thermography, tribology (e.g., lubricating oil analysis), Literature KBM MDSM ML MA

production process parameters, visual inspection, ultrasonic, or [8], [9], [14]    

failure mode (e.g., max rotation speed) analysis. [24], [25], [26], [27], [28], [29]    
Estimating the Remaining Useful Life (RUL) is a central
[16]    
task for optimizing repair intervals and minimizing costs of
[19]    
unscheduled outages created by machine failures. Throughout
years, methods have clearly shifted from rule-based to data- [18]    
driven approaches applying statistical or machine learning [30]    
techniques [24] such as Support Vector Machines, Bayesian [17], [15], [17], [20]    
learning techniques, Hidden Markov models, similarity-based [21], [22], [23] *  
approaches, or unsupervised learning methods. However, most
of these approaches heavily rely on manual investigation and * maturity including maintenance aspects, but not specifically for KBM
identification of possible failure characteristics and subsequent Hence we build on the concept of prescriptive maintenance
feature engineering. model (PriMa) presented in [10] and propose a reference model
In order to overcome manual feature engineering, a number entitled “PriMa-X” that supports the implementation of a
of Deep Learning based methods have been investigated: prescriptive maintenance strategy and the assessment of its
Convolutional Neural Networks have been applied for maturity level, facilitates the integration of data-science
detection faults in rotating machinery [25] and for detecting methods for predicting future events, and identifies action
structural damages [26]. However, both approaches have been fields to reach an enhanced target maturity state and thus higher
evaluated in simulated environments, which shows that Deep prediction accuracy.
Learning is still in its infancy and requires further systematic
research (e.g., standard datasets, insight into black box models, 3. The PriMa-X Reference Model
transferring models, imbalance in training data, etc.) before it
can be applied in the field of prescriptive maintenance [27]. 3.1. Fundamentals and Scope
Anomaly detection, which refers to the problem of finding The PriMa-X reference model (cf. Fig. 2) constitutes a three-
anomalous patterns and outliers in data is a complementary, layer portfolio matrix, covering objectives, challenges as well
frequently used data-driven maintenance technique. Recent as machine learning methods and necessary IT-infrastructures
approaches [28, 29] apply Long-Short-Term-Memory based for realizing a prescriptive maintenance strategy within an
algorithms for that purpose. However, those existing anomaly existing production system. Each layer of the model is broken
detection approaches operate on machine sensor data only down and in-depth analyzed for each step of the introduced
without taking into account production process parameters. prescriptive maintenance process, namely, 1. Analysis and

This is a resupply of March 2023 as the template used in the publication of the original article contained errors. The content of the article has remained unaffected.
 1042 Tanja Nemeth et al. / Procedia CIRP 72 (2018) 1039–1044

Diagnosis, 2. Prediction Model Building, 3. Prescriptive weaknesses are identified for each prescriptive
Maintenance Decision Support, 4. Maintenance Planning and maintenance process step.
5. Execution and Documentation. 3. Definition of a target state maturity and underlying
action fields: A company-specific target maturity level
and necessary measures for its realization are defined
for each prescriptive maintenance process step and will
be considered as input for the next iteration circle.

The first element is described in detail within the following

sub-section. Elements 2 and 3 are outlined in the final section
“Conclusion and Future Research Agenda” of this paper.
3.2. Planning and implementation of a prescriptive
maintenance strategy
For each prescriptive maintenance process step, the three
layers objectives, challenges and ML methods of the reference
Fig. 2. PriMa-X –Three-Layer Portfolio Matrix model are presented within this section. The focus is hereby
placed on the first layer objectives. The key findings of the
Prescriptive maintenance maturity is assessed for a second and third layer are summarized within the tables 2-6.
maintenance and data analytics dimension. Both dimensions
have decisive influence on the success of a prescriptive 1. Analysis & Diagnosis: The process step “Analysis &
maintenance strategy, due to the strong interdisciplinary Diagnosis” firstly aims at building up ground truth datasets in
application of maintenance and production system order to facilitate derivation and description of patterns in
competencies on the one hand and data analytics expertise on stored data sets, which indicate a failure. For this purpose,
the other hand. historical shop floor data, including product, process and
Before implementing the proposed reference model, a target machine as well as maintenance management and cost data, has
application area for the realization of prescriptive maintenance to be acquired from ERP, MES, quality or maintenance
within a production system has to be selected. It can be a single management systems, condition monitoring and PLC controls.
machine, a production line of serial and/or parallel interlinked Relevant data is aggregated, normalized and stored within a
machine associations or a whole production area, characterized comprehensive data model. An appropriate data space
by high internal criticality (due to far-reaching effects of infrastructure and streaming concept (supporting data batching
downtimes, missing redundancies of machines etc.) and high to structured data streaming) is required to store and analyse
importance (production of runner products etc.). Several datasets with large volume, velocity and variety. Secondly, the
concepts for assessing the equipment priority, such as the derived failure patterns must be inspected systematically in
equipment index proposed in [31], are published in the order to understand their technical characteristics, criticality
scientific literature and therefore not further discussed in this (e.g. downtime, costs and occurrence) and their effects on
paper. maintenance key performance indicators. Thirdly, awareness
of failure depended effects in interrelated functional areas
(maintenance, production planning, quality management) has
to be generated within this process step.
Table 2. Analysis & Diagnosis | Challenges and ML methods
Challenges ML methods
• Missing (digital) data • Building up a ground truth
availability • Prediction problem formalization
• Syntactically, structurally and • Expert knowledge formalization
semantically heterogeneity of
• Raw data collection and
data
Sampling
• Diversity of data infrastructures (e.g. Python, R, Shell Scripts)
• Poor data quality • Data preprocessing (e.g.,
Fig. 3. Scope and elements of PriMa-X
• Lacking data storage capability - cleansing, sampling,
PriMa-X builds on three iterative steps (cf. Fig. 3): volume and complexity of data transformation, aggregation)
1. Planning and implementation of a prescriptive • Systematic requirements • Exploratory data analysis (e.g. R,
elicitation and specification Python)
maintenance strategy: A planned and stepwise
• Accuracy of patterns • Hypothesis formulation
execution of the introduced prescriptive maintenance
process enables a systematic realization of a
prescriptive maintenance strategy. 2. Prediction Model Building: In the second process step,
2. Assessment of the prescriptive maintenance maturity: future machine failures are predicted based on the previously
The current state maturity level is assessed and derived failure patterns and incoming data sets in order to
determine their probability of occurrence. Besides the defined

This is a resupply of March 2023 as the template used in the publication of the original article contained errors. The content of the article has remained unaffected.
 Tanja Nemeth et al. / Procedia CIRP 72 (2018) 1039–1044 1043

input data sets of process step 1, further prediction relevant functional areas (maintenance, production planning, and
input data shall be acquired in order to enhance the prediction quality management).
accuracy. The future production program, which indirectly
reveals future machine loads, can, for example, serve as a Table 5. Maintenance Planning | Challenges and ML methods
valuable additional input parameter for the prediction. As a Challenges ML methods
result, transparency of the expected impact of future machine • Data access to MMS • Data visualization and interactive user
failures (e.g. downtime, maintenance KPIs and costs) is • Fast data processing interfaces (concept maps, dashboards,
generated and mechanical conditions and quality deviations capability networks, etc.)
can be monitored in their current and future state. • Establishment of feedback • Online analytical processing (OLAP)
loops to process the final • Meta-analysis
Table 3. Prediction Model Building | Challenges and ML methods judgement of the user • Similarity-based learning (e.g. case-
Challenges ML methods • High confidence of based reasoning)
planner in the prescribed • Information-based learning (e.g.
• Real-time access to actual • Definition of training, test, and
actions decision trees)
planning data validation datasets
• Fast data processing • Selection of suitable machine
capability learning and optimization algorithms 5. Execution & Documentation: Besides the obvious task of
• Successful and significant • Model building / Training timely executing a maintenance order, the last step “Execution
predictions: • Systematic model evaluation and & Documentation” includes the measurement and control of
• High prediction accuracy validation using standard machine the share of successful predictions and effective
learning quality assessment metrics
• High probability of recommendations as well as the a-posteriori specification and
occurrence • Packaging model as data product
concretization of machine failures due to the operator´s
feedback. This information is used for the semi-automatic
3. Prescriptive Maintenance Decision Support: The main refinement of the machine learning methods and the decision
objective of this step lies in the systematic prioritization and model. Finally, the improvement of maintenance key
prescription of maintenance activities. By defining company performance indicators due to the prescribed and bundled
specific operational (e.g. environmental factors, downtime) and maintenance activities is quantified.
management decision parameters (e.g. risk disposition, budget
targets, strategic prioritization) a dynamic set of decision rules Table 6. Execution & Documentation | Challenges and ML methods
is derived and integrated into a multi criteria decision model. Challenges ML methods
The decision model automatically prioritizes the predicted • Provision of relevant information for • Data product deployment
machine failures of process step 2. Considering the current maintenance execution • Data product scaling and
service plan, availability of spare parts and resources a specific • Integration of comprehensive process real-time streaming
maintenance activity can be prescribed in order to avoid the understanding and feedback of connection
employees into the model • Communication and
machine failure and its associated negative impacts.
• High effectiveness of recommended automatic report
Table 4. Prescriptive Maintenance Decision Support | Challenges and ML actions generation
methods • High confidence of maintainer in • Interactive interfaces
Challenges ML methods prescribed actions • Continuous monitoring of
• Measuring and controlling the number prediction accuracy and
• Fast data processing capability • Application of trained and
of highly effective recommendations effectiveness
• Successful and significant evaluated prediction models
prescriptions of maintenance on live-data
activities: • Visualization and reporting 4. Conclusion and Future Research Agenda
o High accuracy of results of prediction results
• Contextualization of The presented reference model constitutes a comprehensive
o Prioritization of results
prediction results (data approach for systematically realizing a prescriptive
o Interpretability of results
sources, timeliness, etc.) maintenance strategy in both operational and management
• Dynamic nature of operational and
business workflows and related • Expert feedback collection process levels of a production system. Besides, typical
decision preferences and parameters (Active Learning) challenges (e.g. multiplicity of data sources, lacking data
quality, and variety of process interrelations) within each
prescriptive maintenance process step and appropriate ML
4. Maintenance Planning: The fourth process step aims at the methods to overcome these challenges are highlighted. As a
creation of concrete maintenance orders within the company´s result of applying the model, companies benefit from optimized
maintenance management system (MMS). Therefore, domain intervals between repairs, minimized costs of unscheduled
experts manually accept (or reject) the previously prescribed machine breakdowns and prevented unexpected consequential
maintenance activities in order to integrate their implicit impacts on associated functional areas.
knowledge and context sensibility within the prescriptive In order to further enhance an existing prescriptive
maintenance process. By bundling maintenance activities, e.g. maintenance strategy in a targeted way, there is a clear need of
a prescribed maintenance activity and a regular preventive task, prescriptive maintenance maturity assessment, which is one
an additional efficiency improvement on operational and major task on the author´s future research agenda. The concept
management level can be achieved within all interrelated is outlined within the next paragraph.

This is a resupply of March 2023 as the template used in the publication of the original article contained errors. The content of the article has remained unaffected.
 1044 Tanja Nemeth et al. / Procedia CIRP 72 (2018) 1039–1044

We pursue an analytical, rather than an empirical approach [13] Institute of Technology Management. Industry Study 2016 Manufacturing
(based on guided interviews with domain experts) to assess the Data Analytics, University of St. Gallen, 2016, retrieved from:
current state maturity level of each prescriptive maintenance http://tectem.ch/content/2-news/20170401-item-study-manufacturing-
process step. For each prescriptive maintenance process step, data-analytics-2017/item-hsg_manufacturing-data-analytics_general-
relevant and measureable key indicators for the data analytics report.pdf, accessed on: 18.12.2017.
and maintenance dimension are derived, e.g.: [14] Pawellek G. Integrierte Instandhaltung und Ersatzteillogistik:
 Maintenance dimension: maintainability, reliability, Vorgehensweisen, Methoden, Tools. Berlin, Germany: Springer, 2013.
availability, repair- and downtime, cost and human [15] Glawar R, Kemeny Z, Nemeth T, Matyas K, Monostori L, Sihn W. A
resource effectiveness Holistic Approach for Quality Oriented Maintenance Planning Supported
 Data analytics dimension: data quality metrics by Data Mining Methods. Procedia CIRP, 2016;57:259-264.
(structure, information and veracity), accuracy of [16] Aghezzaf EH, Abdelhakim K, Phuoc LT. Optimizing Production and
failure patterns, certainty and reliability of predictions Imperfect Preventive Maintenance Planning‫׳‬s Integration in Failure-Prone
By quantifying the defined key indicators, the current state Manufacturing Systems. Reliability Engineering & System Safety,
maturity level is assessed, and weaknesses are identified. These 2016;145:190-198.
results serve as a basis for the third element of the reference [17] Wang J, Zhang L, Duan L, Gao RX. A New Paradigm of Cloud-based
model: A company specific target maturity state is defined Predictive Maintenance for Intelligent Manufacturing. J. of Intelligent
based on strategic priorities of the company. Due to the Manufacturing, 2016;28(5):1125-1137.
quantified key indicators and a set of prescriptive maintenance [18] Arab A, Ismail N, Lee LS. Maintenance scheduling incorporating dyn-
influencing factors on operational, tactical and strategic level, amics of production system and real-time information from workstations.
action fields to achieve the target state can be specified. J. of Intelligent Manufacturing, 2013;24(4):695-705.
[19] Bärenfänger-Wojciechowski S, Austerjost M, Henke M. Smart Maint-
References enance-Asset Management der Zukunft: Ein integrativer Management-
Ansatz. wt-online, 2017:1-2:102-106.
[1] Bokrantz J, Skoogh A, Berlin C, Stahre J. Maintenance in digitalised
[20] Abramovici M, Gebus P, Göbel JC, Savarino P. Provider-Driven
manufacturing: Delphi-based scenarios for 2030. Int. J. of Production
Knowledge Allocation Concept for Improving Technical Repair Tasks in
Economics, 2015;191:154-169.
IPS 2 Networks. Procedia CIRP, 2017;64:381-386.
[2] Lee J, Bagheri B, Kao HA. A cyber-physical systems architecture for
[21] Muchiri AK, Ikua BW, Muchiri PN, Irungu PK. Development of a
industry 4.0-based manufacturing systems. Manufacturing Letters,
theoretical framework for evaluating maintenance practices. Int. J. of
2015;3:18-23.
System Assurance Engineering and Management, 2017;8(1):198-207.
[3] Fitouri C, Fnaiech N, Varnier C, Fnaiech, F, Zerhouni N. A Decison-
[22] Mehairjan RP, van Hattem M, Djairam D, Smit, JJ. Development and
Making Approach for Job Shop Scheduling with Job Depending
implementation of a maturity model for professionalising maintenance
Degradation and Predictive Maintenance. IFAC-Papers On Line,
management. In Proceedings of the 10th World Congress on Engineering
2016;49:1490-1495.
Asset Management. Springer, Cham, 2016:415-427.
[4] Thoben KD., Wiesner S, Wuest T. (2017). “Industrie 4.0” and Smart
[23] Schumacher A, Erol S, Sihn W. A maturity model for assessing Industry
Manufacturing–A Review of Research Issues and Application Examples.
4.0 readiness and maturity of manufacturing enterprises. Procedia CIRP,
Int. J. of Automation Technology, 2017;11(1).
2016;52:161-166.
[5] Monostori L, Kádár B, Bauernhansl T, Kondoh S, Kumara S, Reinhart G,
[24] Sikorska JZ, Hodkiewicz M, Ma L. Prognostic modelling options for
… Ueda K. Cyber-physical systems in manufacturing. CIRP Annals-
remaining useful life estimation by industry. Mechanical Systems and
Manufacturing Technology, 2016;65(2):621-641.
Signal Processing, 2011;1803-1836.
[6] Roy R, Stark R, Tracht K, Takata S, Mori M. Continuous maintenance and
[25] Janssens O, et al. Convolutional neural network based fault detection for
the future–Foundations and technological challenges. CIRP Annals-
rotating machinery. J. of Sound and Vibration, 2016;377:331-345.
Manufacturing Technology, 2016;65(2):667-688.
[26] Abdeljaber O, Avci O, Kiranyaz S, Gabbouj M, Inman DJ. Real-time vib-
[7] Matyas K, Nemeth T, Kovacs K, Glawar R. A procedural approach for
ration-based structural damage detection using onedimensional convo-
realizing prescriptive maintenance planning in manufacturing industries.
lutional neural networks. J. of Sound and Vibration, 2017;388:154-170.
CIRP Annals-Manufacturing Technology, 2017;66:461-464
[27] Zhao R, Yan R, Chen Z, Mao K, Wang P, Gao RX. Deep Learning and Its
[8] Colledani M, Tolio T, Fischer A, Iung, B, Lanza G, Schmitt R, Vancza J.
Applications to Machine Health Monitoring: A Survey. arXiv,
Design and management of manufacturing systems for production quality.
2016;1612.07640.
CIRP Annals-Manufacturing Technology, 2014;63(2):773-796.
[28] Yadav M, Malhotra P, Vig L, Sriram K, Shroff G. Ode-augmented training
[9] ElMaraghy W, ElMaraghy H, Tomiyama T, Monostori L. Complexity in
improves anomaly detection in sensor data from machines. arXiv,
engineering design and manufacturing. CIRP Annals-Manufacturing
2016;1605.01534.
Technology, 2012;61(2):793-814.
[29] Malhotra, P., TV V, Ramakrishnan A, Anand G, Vig L, Agarwal P, Shroff
[10] Ansari F, Glawar R, Sihn W. Prescriptive Maintenance of CPPS by
G. Multi-sensor prognostics using an unsupervised health index based on
Integrating Multi-modal Data with Dynamic Bayesian Networks. In:
lstm encoder-decoder. arXiv, 2016;1608.06154.
Machine Learning for Cyber Physical Systems, Publisher: Springer.
[30] Settles B. Active learning. Synthesis Lectures on Artificial Intelligence
2017;(In Press).
and Machine Learning; 2012;6.1:1-114.
[11] Khoshafian S, Rostetter C. Digital Prescriptive Maintenance, Internet of
[31] Biedermann H. (Ed.). Smart Maintenance - Intelligente, lernorientierte
Things, Process of Everything, BPM Everywhere, 2015
Instandhaltung. TÜV Rheinland Group, Germany Media, 2015.
[12] Lueth KL, Patsioura C, Williams ZD, Kermani ZZ. Industrial Analytics
2016/2017: The current state of data analytics usage in industrial
companies. Technical Report. IoT Analytics. 2016.

This is a resupply of March 2023 as the template used in the publication of the original article contained errors. The content of the article has remained unaffected.