1.

4
Foundations of Real Time Predictive
Maintenance with Root Cause Analysis

Franz Wotawa1 , David Kaufmann1 , Adil Amukhtar1 , Iulia Nica1 ,

Florian Klück2 , Hermann Felbinger2 , Petr Blaha3 , Matus Kozovsky3 ,
Zdenek Havranek3 and Martin Dosedel3
1 Graz University of Technology, Austria
2 AVL List GmbH, Austria
3 Brno University of Technology CEITEC, Czech Republic

Abstract
Research on cyber-physical systems comes to the fore with the increasing
progress of applications in the field of autonomous systems. Therefore, there
is a growing interest in methods for enhancing reliability, availability, and
self-adaptation of such systems in safety critical situations. Hence, it is
essential that autonomous systems are equipped with a detection system to
observe faulty behaviour in real time or to predict failing operations to avoid
safety critical scenarios, which may harm people. To bring or hold a system
within healthy conditions, not only detecting a faulty behaviour is important,
but also to find the corresponding root cause.
In this article, we introduce different methods which make use
of detecting unexpected behaviour in cyber-physical systems, for the
localization of faults. The first approach, model-based diagnosis uses logic
to represent a cyber-physical system to perform reasoning for computing
diagnosis candidates. A second promising approach deals with simulation-
based diagnosis systems, using digital twin models to produce faulty
behaviour data in advance, and to find correlations with the original cyber-
physical system’s behaviour, for diagnosis. For the third method the focus is
set on artificial intelligence (machine learning and neural networks), where

47
48 Foundations of Real Time Predictive Maintenance

the goal is to utilize a huge amount of health and safety critical observations
of the system for training to approximate the behaviour associated with faulty
and safety critical states.

Keywords: model based diagnosis, model based reasoning, simulation based

diagnosis, digital twin, AI based predictive maintenance, AI based diagnosis,
abstract model, datacentre design, energy efficiency of datacentre, energy
efficient metrics, datacentre carbon footprint computation.

1.4.1 Introduction and Background

Predictive analytics deals with forecasting the future progression of
a situation and has a wide range of applications, including weather
forecasting, epidemiology prediction, stock market prediction, and predictive
maintenance. When implementing predictive maintenance, predictive
modelling plays a major role. It aims to guarantee a robust prediction result,
which can save considerable production downtime and either prevent or
diminish economic loss. Considering information utilization and modelling
mechanism, the predictive modelling techniques can be classified into three
groups: physics-based, data-driven, and model-based.
The physics-based approach describes the physical behaviour of a system
using the first principle as a series of ordinary or partial differential equations
according to the law of physics [1][6]. However, the construction of a physics
model is usually difficult since it requires detailed and complete knowledge
about the system. Still, this kind of model lacks extensive failure samples to
determine the model parameters in practice.
The data-driven approach constructs a model representing the underlying
relationship of a system based on data mining techniques. The data-driven
approach could be grouped into two categories including statistical and
machine learning based methods. The typical statistical method used, include
the autoregressive model and its variations, linear regression, Wiener process,
and Gamma process among others. Machine learning based methods include
algorithms such as artificial neural networks, clustering techniques, extreme
learning machines, fuzzy logic, and deep learning models. However, the
performance of the data-driven model is sensitive to the size and quality
of the collected dataset. It is important to note that data-driven models are
extremely domain specific. Therefore, the selection of such models is a
crucial part of the process.
 1.4.1 Introduction and Background 49

The model-based approach takes advantage of established physical

knowledge and collected data to enhance the prediction performance. It
typically involves two steps including model construction and model updating
[2]. First, analytical models are built based on the physical or empirical
model representing the situation evolving in a quantitative manner. These
models are then updated with newly acquired information to predict the future
progression of the situation based on inference. Comparing with the data-
driven approach, the model-based approach requires less historical data to
construct the models. The predicted value is associated with a confidence
level, resulting from the uncertainty involved in the prediction process [3].
Over the past 30 years, predictive maintenance has been evolving from
predicting failures based on periodic visual inspections to continuous real-
time monitoring of assets and external data with alerts based on statistical
techniques such as regression analysis for at least one important asset.
Furthermore, the advent of Industrial Internet of Things (IIoT) technology
has significantly optimized industrial operations management by connecting
industrial assets with information systems and, hence, with business
processes. Predictive Maintenance 4.0 (PdM 4.0) or simply Maintenance 4.0,
is among the major focus points of IIoT. In [4] the authors identify four levels
of maturity in predictive maintenance, depicted in Figure 1.4.1.
Many companies are combining the capabilities of IIoT and Big Data
to predict equipment malfunctions. The accuracy of the forecast is further

Figure 1.4.1 Four levels of maturity in predictive maintenance.

50 Foundations of Real Time Predictive Maintenance

getting more precise with improved Artificial Intelligence (AI) techniques

and machine learning tools.
As depicted in [5], Maintenance 4.0 forms a subset of smart manufact-
uring systems which are autonomous in their operation, capable of predicting
failures and triggering maintenance activities. These systems consist of
smart equipment in form of embedded or cyber-physical systems form-
ing the digital twin of physical assets. To achieve near-zero defects,
near-zero downtime and automated decision making based on condition
monitoring, top diagnosis and prognosis techniques need to be implemented.
Finally, the most advanced form of maintenance is prescriptive
maintenance which builds on PdM and provides further guidance on the
maintenance task, including diagnosis capabilities. Prescriptive maintenance
strategies extensively use advanced data processing and visualization
techniques such as graph analysis, simulations, neural networks, complex
event processing, heuristics, and machine learning. These tools can calculate
the timing and the effect of failure, thus, deciding on the priority and urgency
of the maintenance activity.
In Figure 1.4.2, we depict a simplified system architecture, showing how
the different approaches contribute to diagnosis of systems. Simulation-based
diagnosis as well as AI-based diagnosis, utilize models that are obtained in a
pre-offline phase, depicted on the right. Model-based diagnosis makes use of

Figure 1.4.2 Diagnosis system architecture.

 1.4.2 Foundations 51

mainly abstract models for diagnosis directly not requiring an offline phase.
In the following, we discuss the different approaches and their foundations in
more detail.

1.4.2 Foundations
In the following, we discuss the foundations behind the diagnosis, i.e.,
the detection of failures and the identification of its root causes in the
context of predictive maintenance. In particular, we focus on methods from
artificial intelligence considering model-based diagnosis, machine learning,
and specifically neural networks. Instead of a detailed discussion of the
foundations, we briefly introduce underlying ideas and provide references to
related literature for the interested reader.

1.4.2.1 Model-based Diagnosis

Model-based diagnosis or reasoning from the first principle has been
developed in the 80s of the last century as an answer to challenges arising
when using logic reasoning as a basis for applications like configuration
and decision support. Instead of formalizing the knowledge-base in a way
from observations to causes such that diagnosis can be directly derived using
ordinary deduction, the idea was to formalize knowledge either in form of
relations or as rules where causes imply their effects. Instead of deduction
abduction or in a more general setting non-monotonic reasoning was used as
an underlying reasoning mechanism (see [19], [20], [21]).
The idea behind model-based diagnosis is to take a model of a system,
which is usually called a system description SD, and observations OBS
for diagnosis computation. In this setup, SD comprises the structure of
the system comprising interconnected components, and the behaviour of
the components. For the latter, we explicitly introduce health states for
components like working abnormally (ab), or correctly (i.e., not abnormal
(¬ab)). For example, the correct behaviour of components can be formalized
using an implication, i.e., ¬ab(C) → behav(C), where C is a component and
behav(C) the behaviour of C. Whenever the component works as expected
the behaviour is determined. However, if we assume C to be wrong, the
implication does not allow us to determine behaviour. Hence, the component
may work appropriately even when considered to be faulty. Note that this
modelling allows also to specify a behaviour for any incorrect health state if
required.
52 Foundations of Real Time Predictive Maintenance

When assuming a model of a system comprising components COM P

and observations, we are able to compute diagnoses. Informally, a diagnosis
explains a faulty behaviour. In the case of model-based diagnosis, we
are interested in assigning a health state to every component in COM P
such that the observations are not in contradiction with the components’
behaviour. Hence, diagnosis becomes searching for health states. In order
to be applicable in practice, diagnosis reasoning often utilizes simplifications
like searching only for diagnoses where one component is considered to be
faulty, and all others are working as expected. Alternatively, diagnosis search
may focus on parsimonious diagnoses, i.e., health assignments to components
unequal to ¬ab, where we are not able to switch a component from being
faulty to working correctly.
Model-based diagnosis computation in general is hard and requires a lot
of computational resources. However, considering today’s hardware, most
recent algorithms, and the availability of fast theorem provers, diagnosis can
be computed within a reasonable amount of time, i.e., within a fraction of a
second even for larger systems (see [22]). For a more detailed discussion on
model-based diagnosis, modelling, formal definitions, and its application to
self-adaptive systems we refer to [23] and most recently [24].

1.4.2.2 Machine Learning Based Diagnosis

Machine learning algorithms have shown promising solutions and improved
decision-making processes by analysing an enormous amount of data. The
use of these algorithms has grown rapidly in the recent years which
helps systems to act intelligently without being explicitly programmed [7].
Machine learning techniques are often used to detect faulty behaviours of the
system [8], [9]. For example, [10] used Support Vector Machine (SVM), a
machine learning algorithm to model linear and non-linear relationships, to
model 9 fault states of the modular production system with different kernel
functions namely Sigmoid, RBF, polynomial and linear kernel functions. The
work presented a 100% classification rate on all kernel functions except for
the sigmoid kernel (52.08% classification rate).
Machine learning algorithms are mainly divided into four categories
explained below:
• Supervised: This type of learning typically learns a function based on the
sample input and output pairs. The goal of the function is to classify/map
a new input instance to the respective output [11]. Please note that the
data samples provided during the training are labelled.
 1.4.2 Foundations 53

• Unsupervised: Unsupervised learning involves understanding the

distribution of the data given the data is unlabelled [11]. These types
of algorithms are mostly used for feature generation, dimensionality
reduction, extracting hidden patterns, clustering/grouping data points,
and exploratory analysis.
• Semi-Supervised: Data points could be rarely labelled in real world
[12]. For example, in the fraud detection problem, there could be few
occurrences of fraud transaction leaving too much non-fraud detection
data. Thus, semi-supervised learning comes into play by generating new
instances from the less seen (minatory output), often called synthetic
data generation. It’s a hybridization of “supervised” and “unsupervised”
where the goal is to model better predictions given the data is highly
unlabelled.
• Reinforcement: Reinforcement learning is an area of machine learning
in which an agent is trained to learn the optimal behaviour for a given
environment [13]. The goal of reinforcement learning is to find the
best possible actions such that reward is maximized and the risk is
minimized. Reinforcement learning is mostly useful for automation e.g.,
autonomous driving.
Based on the application, nature of the data and learning outcome, various
machine learning algorithms can be chosen for fault diagnosis in complex
systems. For this case study, we model the fault diagnosis problem with one
of the supervised machine learning algorithms called Bootstrap Aggregation
(Bagging).

1.4.2.3 Artificial Neural Networks for Diagnostics

Machine learning as well as deep learning techniques are very popular in
many areas of engineer’s work. The connection of the AI approach and
technical diagnostics especially in the field of predictive maintenance of
machines [14] is a very actual problem and directly addresses the Internet of
Things as well as Industry 4.0 topics [15]. Big data processing algorithms,
necessary for modern AI techniques application, are overviewed in, e.g.,
[16], standard machine learning approaches, mostly containing statistical
algorithms [17] like SVM, k-NN, PCA, Mahalanobis-Taguchi strategy etc.,
are commonly used, but mainly using of powerful and very popular neural
networks is currently growing. There exists a lot of NNs types used for
diagnostics of the machines, but the convolutional neural network is one of
the most recommended and also used types [18]. Mostly, NN algorithms
54 Foundations of Real Time Predictive Maintenance

run on the dedicated and powerful hardware designed especially for such
purposes.
The shift from cloud AI processing to local intelligence architecture is
described in [25]. According to that paper, AI has a strong potential for
sensor solutions in the future. Reasons are the increasing complexity of
sensors, the increasing amount of generated raw data, and the requirement for
straightforward data fusion from several sensors. The integration of wireless
communication capabilities in smart sensors makes them usable also as an
IoT device [26]. This process must be accompanied by the integration of
safety- and privacy-aware functions.

1.4.3 Related Research

Predictive analytics intends to make predictions about future progressions,
based on domain knowledge and historic data combined with physic-based,
model-based or machine-learning modelling techniques. In the context of
predictive maintenance (PdM), predictive modelling is used for failure
prediction and prescription of operation and maintenance strategies. Here,
the main objective is to obtain accurate and robust prediction results to
avoid unexpected system downtime. Predictive maintenance is a condition-
driven maintenance program that monitors the mechanical condition, system
efficiency, and other indicators to determine the system’s actual mean-time-
to-failure or loss of efficiency. Considering the definition from [27], the
three key steps of a PdM program are data acquisition to obtain data
relevant to system health, data processing to handle and analyse the data or
signals collected and maintenance decision-making to recommend efficient
maintenance actions or adoptions of the operation strategy. Techniques for
maintenance decision support in a PdM program can be divided into two main
categories [27]: diagnostics and prognostics. Fault diagnostics focuses on
detection, isolation, and identification of faults when they occur. In contrast,
prognostics attempts to predict faults or failures before they occur. Jardine
[28] reviewed and compared several commonly used PdM decision strategies
such as trend analysis that is rooted in statistical process control (SPC), expert
systems (ESs), and neural networks. Wang and Sharp [29] discussed the
decision aspect of PdM and reviewed the recent development in modelling
PdM decision support.
Various model-based diagnosis approaches have been applied to fault
diagnosis of a variety of mechanical systems such as gearboxes [30][31],
bearings [32][33][34], rotors [35][36] and cutting tools [37]. Hansen et al.
 1.4.4 Conclusion 55

[38] proposed an approach to more robust diagnosis based on the fusion

of sensor-based and model-based information. Vania and Pennacchi [39]
developed some methods to measure the accuracy of the results obtained with
model-based techniques aimed to identify faults in rotating machines. Two
practical successful applications of maintenance programs using model-based
approaches are: (i) an integrated framework for on-board fault diagnosis
and failure prognosis of a helicopter transmission component and (ii) the
TIGER system [40] that combines several artificial intelligence technologies,
including qualitative model-based reasoning to perform condition monitoring
of gas turbines. Here, the diagnostic mechanism is based on a fault manager
and the three independent tools KHEOPS [67], IxTeT [40] and CA-EN [69].
KHEOPS [41] is a high-speed rule-based system, used to express diagnostic
rules in a classic rule-based formalism and allows the user to set pre-alarm
limits for each parameter. IxTeT [40] is used to either describe the normal
causal reaction or look for specific patterns resulting from known faults. CA-
EN [42] is a model-based supervision system devoted to complex dynamic
systems. CA-EN’s representation formalism allows one to combine empirical
causal knowledge and first principles of the domain.
The effectiveness of predictive maintenance depends on practical factors
such as required planning time and implementation effort but especially
on the achievable quality of condition monitoring, the behaviour of the
deterioration process and system specific fault severity. For instance,
vibration and oil debris monitoring is limited by the accuracy of the
measuring instruments and can therefore be considered as imperfect [52].
In many cases, the imperfect condition information has been combined
with deterioration processes, which were modelled as continuous stochastic
processes. Kallen and Van Noortwijk [43] use a gamma deterioration process,
Peng and Tseng [44] a linear trend with random coefficient plus a Brownian
motion as a second random effect, Ye et al. [45] a Wiener process with
positive drift, and Zio and Compare [46] a Randomized Paris-Erdogan fatigue
crack growth model. Nevertheless, also here inspections have to be performed
in order to obtain condition information. Given the effort and short comings,
PdM should only be applied if the expected benefit outweighs the efforts and
costs during the entire life cycle [47][48][49].

1.4.4 Conclusion
Predictive maintenance mechanisms are the major key to improve the
availability, reliability and safety of cyber-physical systems in relation to
56 Foundations of Real Time Predictive Maintenance

finding or predicting an unexpected behaviour before downtime, defects

or harm to the environment occurs. In this article, we focus on different
approaches for diagnosis, discuss their foundations, and also related research.
However, there remains the questions which diagnosis methods to use and
how to implement them to interact with a specific cyber-physical system. We
elaborate on use cases in two separated articles of this book to answer these
questions.
In these articles, we decided to focus on different diagnosis approaches
based on two systems, a simplified DC e-motor model and a dual three-phase
permanent magnet synchronous motor supported with detailed acausal e-
motor model with the capability of fault injection. The use of model based
and machine learning based approaches is demonstrated on a simplified
DC e-motor model in the article “Real-Time Predictive Maintenance -
Model Based and Machine Learning Based Diagnosis”. The artificial neural
network approaches are demonstrated on a dual three-phase motor diagnosis
and on a diagnosis using smart vibration sensor which is described in
article “Real-Time Predictive Maintenance – Artificial Neural Network Based
Diagnosis”.
In the mentioned articles, we discuss the applicability of diagnosis
algorithms in real-time simulation environments by highlighting a specific
case of how to implement the methods and perform diagnoses on unexpected
behaviour. We obtained promising results encouraging for further research
on the described diagnosis methods depending on the desired detection
dimension, available resources, and model specifications. In addition, the
diagnosis methods deliver the root cause affects which builds the basis for
the research in self-adapting or self-healing systems to bring a system to a
safe state if an unexpected behaviour is detected.

Acknowledgements
This work is conducted under the framework of the ECSEL AI4DI
“Artificial Intelligence for Digitising Industry” project. The project has
received funding from the ECSEL Joint Undertaking (JU) under grant
agreement No 826060. The JU receives support from the European Union’s
Horizon 2020 research and innovation programme and Germany, Austria,
Czech Republic, Italy, Latvia, Belgium, Lithuania, France, Greece, Finland,
Norway. The work was co-funded by grants of Ministry of Education,
Youth and Sports of the Czech Republic, by the Austrian Federal Ministry
of Transport, Innovation and Technology (BMVIT) under the program
 References 57

“ICT of the Future” between May 2019 and April 2022 (more information
can be retrieved from https://iktderzukunft.at/en/). The work was also
supported by the infrastructure of RICAIP that has received funding from
the European Union’s Horizon 2020 research and innovation programme
under grant agreement No 857306 and from Ministry of Education, Youth
and Sports under OP RDE grant agreement No CZ.02.1.01/0.0/0.0/17_043/
0010085.

References
[1] Wang, J., Liang, Y., Zheng, Y., Gao, R. X., and Zhang, F., “An
integrated fault diagnosis and prognosis approach for predictive
maintenance of wind turbine bearing with limited samples,” Renewable
Energy, vol. 145, pp. 642âĂŞ650, 2020. Available online at:
https://www.sciencedirect.com/science/article/pii/S0960148119309371
[2] Song, Z., Zhang, Z., Jiang, Y., and Zhu, J., “Wind turbine
health state monitoring based on a bayesian data-driven approach,”
Renewable Energy, vol. 125, pp. 172âĂŞ181, 2018. Available online at:
https://www.sciencedirect.com/science/article/pii/S0960148118302404
[3] Wang, J., Wang, P., and Gao, R., “Enhanced particle filter for tool wear
prediction,” 2015.
[4] Sept. 2018, SURVEY Predictive Maintenance 4.0 Beyond the hype:
PdM 4.0 delivers results, Mainnovation.
[5] Towards an open-standards based framework for achieving condition-
based predictive maintenance, Kaur, Karamjit & Selway, Matt &
Grossmann, Georg & Stumptner, Markus & Johnston, Alan, (2018)
Proceedings of the 8th International Conference on the Internet of
Things.
[6] Peng, Y., Dong, M., Zuo, M., “Current status of machine prognostics in
condition-based maintenance: a review,” 2010.
[7] I. a. F. M. a. N. R. Sarker, “AI-Driven Cybersecurity: An Overview,
Security Intelligence Modeling and Research Directions,” SN Computer
Science, vol. 2, 2021.
[8] T. a. B. M. a. P. V. Ademujimi, “A Review of Current Machine Learning
Techniques Used in Manufacturing Diagnosis,” pp. 407-415, 2017.
[9] M. a. D. F. a. K. M. a. M. O. Barakat, “Self adaptive growing neural
network classifier for faults detection and diagnosis,” Neurocomputing,
vol. 74, pp. 3865-3876, 2011.
58 Foundations of Real Time Predictive Maintenance

[10] M. Demetgul, “Fault diagnosis on production systems with support

vector machine and decision trees algorithms,” The International Journal
of Advanced Manufacturing Technology, vol. 67, 2012.
[11] J. a. K. M. a. P. J. Han, Data mining concepts and techniques, third
edition, 2012.
[12] M. a. K. M. a. B. E. Mohammed, Machine Learning: Algorithms and
Applications, 2016.
[13] L. P. a. L. M. L. a. M. A. W. Kaelbling, “Reinforcement learning: A
survey,” Journal of Artificial Intelligence Research, vol. 4, pp. 237-285,
1996.
[14] R. Liu, B. Yang and E. Zio: Artificial intelligence for fault diagnosis
of rotating machinery: A review. Mech. Syst. Signal Process., vol. 108,
aug 2018: p. 33–47, ISSN 10961216, doi:10.1016/j.ymssp.2018.02.016.
URL https://linkinghub.elsevier.com/retrieve/pii/S0888327018300748.
[15] A. Chen, F. H. Liu and S. D. Wang: Data reduction for real-time
bridge vibration data on edge. In Proc. - 2019 IEEE Int. Conf. Data
Sci. Adv. Anal. DSAA 2019, Institute of Electrical and Electronics
Engineers Inc., oct 2019, ISBN 9781728144931, p. 602–603, doi:
10.1109/DSAA.2019.00077.
[16] S. Yin, X. Li and H. Gao: Data-based techniques focused on modern
industry: An overview. IEEE Trans. Ind. Electron., vol. 62, issue. 1,
2015: s. 657–667, ISSN 02780046, doi:10.1109/TIE.2014.2308133.
URL https://ieeexplore.ieee.org/document/6748057/.
[17] S. Zhang, S. Zhang and B. Wang: Deep Learning Algorithms
for Bearing Fault Diagnostics - A Comprehensive Review. IEEE
Access, vol. 8, jan 2020: p. 29857–29881, ISSN 21693536,
doi:10.1109/ACCESS.2020.2972859, 1901.08247. URL http://arxiv.or
g/abs/1901.08247
[18] H. Qiao, T. Wan and P. Wang: An Adaptive Weighted
Multiscale Convolutional Neural Network for Rotating Machinery
Fault Diagnosis under Variable Operating Conditions. IEEE
Access, aug 2019: p. 118954–118964, ISSN 21693536,
doi:10.1109/ACCESS.2019.2936625.
[19] Davis, R.: Diagnostic reasoning based on structure and behavior.
Artificial Intelligence 24, pp. 347–410, 1984.
[20] Reiter, R.: A theory of diagnosis from first principles. Artificial
Intelligence 32(1), pp. 57–95, 1987.
[21] de Kleer, J., Mackworth, A.K., Reiter, R.: Characterizing diagnosis and
systems. Artificial Intelligence 56, 1992.
 References 59

[22] Kaufmann, D., Nica, I., Wotawa, F.: Intelligent agents diagnostics
- enhancing cyber-physical systems with self-diagnostic capabilities.
Advanced Intelligent Systems, 2021. DOI https://doi.org/10.1002/ai
sy.202000218.
[23] Wotawa, F.: Reasoning from first principles for self-adaptive and
autonomous systems. In: E. Lughofer, M. Sayed-Mouchaweh (eds.)
Predictive Maintenance in Dynamic Systems - Advanced Methods,
Decision Support Tools and Real-World Applications. Springer (2019).
DOI 10.1007/978-3-030-05645-2.
[24] Wotawa, F.: Using model-based reasoning for self-adaptive control
of smart battery systems. In: M. Sayed-Mouchaweh (ed.) Artificial
Intelligence Techniques for a Scalable Energy Transition – Advanced
Methods, Digital Technologies, Decision Support Tools, and
Applications. Springer (2020).
[25] P. Jantscher: AI In Sensors For IoT [online] Cited 15.6.2021 Available
from: https://siliconsemiconductor.net/article/106227/AI_In_Sensors
_For_IoT.
[26] S. C. Mukhopadhyay, S. K. S. Tyagi, N. K. Suryadevara, V. Piuri, F.
Scotti and S. Zeadally: Artificial Intelligence-based Sensors for Next
Generation IoT Applications: A Review, in IEEE Sensors Journal, doi:
10.1109/JSEN.2021.3055618.
[27] A review on machinery diagnostics and prognostics implementing
condition-based maintenance, Jardine, Andrew & Lin, Daming
& Banjevic, Dragan, Mechanical Systems and Signal Processing,
2006.
[28] A.K.S. Jardine, Optimizing condition based maintenance decisions, in:
Proceedings of the Annual Reliability and Maintainability Symposium,
2002, pp. 90–97.
[29] W. Wang, J. Sharp, Modelling condition-based maintenance decision
support, in: Condition Monitoring: Engineering the Practice, Bury St
Edmunds, 2002, pp. 79–98.
[30] I. Howard, S. Jia, J. Wang, The dynamic modelling of a spur gear in
mesh including friction and a crack, Mechanical Systems and Signal
Processing 15 (2001) 831–838.
[31] W.Y. Wang, Towards dynamic model-based prognostics for transmission
gears, in: Component and Systems Diagnostics, Prognostics, and Health
Management II, vol. 4733, Bellingham, 2002, pp. 157–167.
60 Foundations of Real Time Predictive Maintenance

[32] D.C. Baillie, J. Mathew, Nonlinear model-based fault diagnosis of

bearings, in: Proceedings of an International Conference on Condition
Monitoring, Swansea, UK, 1994, pp. 241–252.
[33] K.A. Loparo, M.L. Adams, W. Lin, M.F. Abdel-Magied, N. Afshari,
Fault detection and diagnosis of rotating machinery, IEEE Transactions
on Industrial Electronics 47 (2000) 1005–1014.
[34] K.A. Loparo, A.H. Falah, M.L. Adams, Model-based fault detection
and diagnosis in rotating machinery, in: Proceedings of the Tenth
International Congress on Sound and Vibration, Stockholm, Sweden,
2003, pp. 1299–1306.
[35] A.S. Sekhar, Model-based identification of two cracks in a rotor system,
Mechanical Systems and Signal Processing 18 (2004) 977–983.
[36] G.H. Choi, G.S. Choi, Application of minimum cross entropy to model-
based monitoring in diamond turning, Mechanical Systems and Signal
Processing 10 (1996) 615–631.
[37] W. Bartelmus, Mathematical modelling and computer simulations as an
aid to gearbox diagnostics, Mechanical Systems and Signal Processing
15 (2001) 855–871.
[38] R.J. Hansen, D.L. Hall, S.K. Kurtz, A new approach to the challenge
of machinery prognostics, Journal of Engineering for Gas Turbines and
Power 117 (1995) 320–325.
[39] A. Vania, P. Pennacchi, Experimental and theoretical application of fault
identification measures of accuracy in rotating machine diagnostics,
Mechanical Systems and Signal Processing 18 (2004) 329–352.
[40] Dousson, C., Gaborit, P., & Ghallab, M. ‘Situation recognition:
representation and algorithms’, Proc. 13th IJCAI, Chambery, France,
1993.
[41] Ghallab, M., & Philippe, H. ‘A Compiler for Real-Time Knowledge-
Based Systems’. Proc. IEEE International Symposium on AI for
Industrial Applications, 287-293, 1988.
[42] Bousson, K., & Trave-Massuyes, L. ‘Fuzzy Causal Simulation in
Process Engineering’. IJCAI-93, Chambery, France. August-September
1993.
[43] Kallen MJ, Van Noortwijk JM. Optimal maintenance decisions under
imperfect inspection. Reliab Eng Syst Saf 2005;90(2–3):177–85.
[44] Peng C-Y, Tseng S-T. Mis-specification analysis in linear degradation
models. IEEE Trans Reliab 2009;58(3):444–55.
 References 61

[45] Ye Z-S, Wang Y, Tsui K-L, Pecht M. Degradation data analysis

using wiener processes with measurement errors. IEEE Trans Reliab
2013;62(4):772–80.
[46] Zio E, Compare M. Evaluating maintenance policies by quantitative
modeling and analysis. Reliab Eng Syst Saf 2013;109:53–65.
[47] Maintenance management of wind power systems using condition
monitoring systems – life cycle cost analysis for two case studies,
Nilsson J, Bertling L., 2007.
[48] Asset life cycle management: towards improving physical asset
performance in the process industry, Schuman CA, Brent AC, 2005
[49] Two probabilistic life-cycle maintenance models for deteriorating civil
infrastructures, Van Noortwijk JM, Frangopol DM, 2004.
[50] Wang W, Christer AH. Towards a general condition based maintenance
model for a stochastic dynamic system. J Oper Res Soc 2000;51(2):
145–55.