IOP Conference Series: Materials Science and Engineering

PAPER • OPEN ACCESS

Intelligent System for Gearbox Fault Detection & Diagnosis Based on

Vibration Analysis using Bayesian Networks
To cite this article: Dedik Romahadi et al 2019 IOP Conf. Ser.: Mater. Sci. Eng. 694 012001

View the article online for updates and enhancements.

This content was downloaded from IP address 173.211.126.40 on 22/11/2019 at 03:44

ISAIME 2019 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 694 (2019) 012001 doi:10.1088/1757-899X/694/1/012001

Intelligent System for Gearbox Fault Detection & Diagnosis

Based on Vibration Analysis using Bayesian Networks

Dedik Romahadi1, Hui Xiong2 and Hadi Pranoto1

1
Mechanical Engineering, Mercu Buana University, Indonesia
2
Mechanical Engineering, Beijing Institute of Technology, China

E-mail: [email protected]

Abstract. Moving energy from one machine to another and functioning to reduce speed while
increasing torque is the ability of the gearbox. Due to many components and the structure
between the components is fairly complex, thus to be able to detect the initial damage,
sophisticated methods is needed. Vibration analysis is a method that has been effective in
detecting the initial damage that occurs in machine. But it takes time and costs are not small to
implement. The purpose of this study is to create an intelligent system capable of detecting
gearbox damage based on data obtained from vibration measurements. Merging of two methods
of vibration analysis and Bayesian networks is done to be able to design the system with the
expected results. A series of multistate nodes are applied to the network and a system review is
performed. Results are given and compared with results provided by the manual analysis. The
results indicate that the system is feasible and reasonable which can assist inidentifying gearbox
damages. This study definitively answers the problem of how to design an expert system capable
of replacing the work of an expert on vibration analysis services.

1. Introduction
Almost all industries use gearboxes to carry out a series of production, gearboxes may operate under
varying load conditions. In industrial applications, gearboxes are a very important component. Gearbox
must work optimally to be able to complete the task as planned. Termination of the production process
and other undesirable things can occur due to gearbox damage, even human casualties. Therefore,
damage to the gearbox needs to be known optimally as early as possible to ensure the operation of an
efficient mechanical transmission system and prevent accidents [1,2].
One condition monitoring technique that are reliable, easy to do and is applied in real life is vibration
analysis. Vibration frequency characteristics are able to show most types of damage that occurs on a
rotating machine and hence, analysis of vibration characteristics useful to identify most types of fault.
Vibration analysis is an attempt to minimize damage to the engine components propagate to other engine
components. Vibration analysis also helps to know the structure of the machine, whether it is still in
stable condition, in accordance with the standard or vice versa. The latest technique commonly used to
diagnose gearbox is by analysing vibration signals taken from the gearbox casing. The aim is to detect
the location and type of damage at the initial stage of development and monitor the condition, to estimate
the lifetime of the machine and prepare appropriate treatment. Gear Mesh Frequency (GMF) and its
repetitions or harmonics and modulation phenomena that produce sidebands are the main components
in the gearbox domain frequency (spectrum). Disturbance conditions can be observed from increases in

Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
ISAIME 2019 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 694 (2019) 012001 doi:10.1088/1757-899X/694/1/012001

the number and amplitude of sidebands. In addition, the source of damage related to the sidebands
distance [3].
Not all engineers know how to analyse vibration on industrial machines, especially on the gearbox.
To analyse the vibrations of the gearbox requires a long learning time and experience in analysing using
standard software from vibration equipment manufacturers. Usually in this case the company must pay
dearly for expert staff services to measure and analyse the condition of the gearbox using vibration data.
Even companies that already have a vibration analyser still have to call and pay for expensive because
they do not understand how to analyse vibration data taken from gearbox. In addition, an expert staff in
vibration analysis also still needs time to analyse spectrum data. Therefore, in solving these problems
the authors propose to create a system that can read the spectrum of vibration data specifically to
diagnose gearbox. A system that requires only input spectrum data and specifications from the gearbox.
Furthermore, the system can provide important information about the gearbox condition, covering where
the point of damage, what actions are needed and how much damage level.
The medical domain has inspired the concept of diagnosis. At present, the types of machine failures
can be identified as root causes with sophisticated diagnostic methodologies. However, the diagnosis is
made after the damage has occurred, including in the reactive action. In fact, observing certain machine
conditions has a problem fault diagnosis in detecting the characteristics of damage that might be hidden
in the dataset. Besides, detecting various damage to the rotating machine is suitable using monitoring
vibration conditions. Signal processing, data acquisition, extraction of other features and features to
reduce steps if required and the last correct classification step that can be obtained from historical data,
expert insight and physical formation, all of which are a common framework for damage diagnosis
systems [4, 5]. BN is able to make the possibilities that occur in the future very well, BN is able to
handle uncertain causal relationships, update probabilities, multi-state variables, carry out two-way
interpretations and handle the lack of data [6, 7, 8, 9]. In various scientific fields, BNs are widely used.
BN's application practices in life are very widely used. BN has long been an interesting topic and
continues to this day, supported by the rapid development of computer science, especially in data mining
and machine learning [10]. Uncertainty arises and become the main component when designing
knowledge-based reasoning and decision makers methods [11]. Related to expert system in
manufacturing, Kobbacy and McNaught et al. emphasize that BNs are most useful when dealing with
uncertainty [12]. BN is a reliable tool for developing expert systems in the field of artificial intelligence.
It is advantageous in reflecting and diagnosing intricate systems with incomplete, uncertain and even
conflicting information. The probabilistic graphical model of BN shows a probabilistic dependency
relationship in a variable connected via an acyclic graph. Pearl introduced in the early 1980s [13, 14],
in probabilistic inference and the realm of knowledge discovery has been applied successfully [15].
Previous techniques of artificial intelligence systems for damage detection use fault tree analysis
(FTA), It is an analytical tool that graphically translates combinations of errors that cause system
failures. This technique is useful for describing and assessing events in the system. This FTA method is
effective in finding the core of the problem because it ensures that an unwanted event or the loss caused
does not originate at one point of failure. FTA identifies the relationship between causal factors and is
displayed in the form of a fault tree involving a simple logic gate [16, 17]. The advantage of using the
FTA method because it has a clear, logical, intuitive, and usable system compared to some other systems.
Nevertheless, the computation of the probability of failure becomes intricate when the structure of a
Fault Tree is large, therefore an uncertain causal relationship and a multi-state variable cannot be done
with an FTA [18, 19, 20], the component fault and reasoning generally do not pay attention to frequency
failures and relationships for different components of the system. Calculation and analysis assuming
that failures for independent components or subsystems can be subject to major errors. Regarding
intelligent fault inference system for rotating machine and the methods used in this study was adopted
from his paper, Bin Gang Xu used system fault assessment is structured by the BN concept, makes
several failure sequences and the relationship between component failure with the BN model and
validated fault assessment estimates. The proposed network applications in a flexible rotor’s vibration
fault diagnosis showed strong inferential abilities. Various error inferences can be continued, related to

2
ISAIME 2019 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 694 (2019) 012001 doi:10.1088/1757-899X/694/1/012001

a single error, a combination of maximum errors and maximum network status, including with
reasonable probabilistic information [21]. Developing BNs applications for rotating machines, this study
uses spectrum frequencies as input data to identify symptoms of damage that occurs in the gearbox.

2. Methodology
2.1. Bayesian networks
Bayesian Inference and Bayesian network (BN) are decompositions of the Bayes’ theorem. The
development of the Bayes theorem began in the 1760s, which new information to update possibilities.
Some previous statisticians, for instance Pierre-Simon La-Place, decision-making methods and
systematic statistical inference are developed from the Bayes formula [22, 23]. A BN was proposed by
Pearl in 1988, which is the method that continues to be discussed today [24]. A special method called
Bayesian Analysis that uses previous statistical information and success in many practical
implementations.
Bayes’ Theorem:
P(M | N ) P (N )
P(N | M ) = (1)
P(N)
Building the BN, a system is constructed with a statistical technique known as the Bayes theorem
namely Conditional Probability. Calculation of the probability of an event N if known event M has
occurred is called Conditional Probability, represented by P(M), as shown in equation 1. This theorem
is used to calculate the odds of a data set to enter into a particular class based on inference existing data.
The combined probability factoring can be obtained by the regulation rule of probability theory, as
shown in equation 2.
P ( x1 , x2 , x3 , , xn ) =  P  xi | Parents ( xi ) (2)
i

A BN is a Probabilistic one simple graphical model constructed from theory probability and graph
theory. Probabilistic theory is related directly with data, while graph theory is related immediately with
the representation you want to get. As for example, a BN can represent relationships between disease
and symptoms. BN is able to calculate the probability of various symptoms of the disease. Building BN,
an expert is needed to apply a concept of knowledge to the network. Knowing how a domain and its
uncertainty can be applied in the BN and see how to use BN for domain needs. When observing the
values of several variables and condition them in new information. This conditioning process is called
probability inference that is carried out through the network. The information flow on the BN is not
limited to the direction of the arc. In the probabilistic network, this becomes the task of calculating the
posterior probability distribution for a set of request nodes, some evidence nodes provide values [24,
26].
Make probabilistic inference is the ability of the BN. Probabilistic inference is predicting the grade
of an element that can’t be known immediately by using the values of other known variables. An
example of probabilistic inference is determining the conditional probability of the type of gearbox
damage based on the symptoms of damage if it is known that the frequency spectrum characteristics of
damage occur. Probabilistic inference can be done if first obtained Joint Probability Distribution (JPD)
of all the modelled variables. JPD is the probability of all variable events occurring simultaneously.
Probabilistic inference can be done if the BN structure has been built. In diagnosing gearbox damage,
the relationship between variables and the probability of variable values is unknown. The construction
of BN from the data consists of structural construction or also called the qualitative stage, which is
looking for connectivity between the modelled variables and parameter estimation or also called the
quantitative stage, which is calculating probability values.

2.2. Vibration signature analysis

Analysis of rotating machine damage usually uses vibration signals in the form of spectrum frequencies
obtained from Fast Fourier Transform (FFT). This method is a reliable tool for analyzing the location

3
ISAIME 2019 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 694 (2019) 012001 doi:10.1088/1757-899X/694/1/012001

of damage. FFT combined with vibration level values provides information about the condition of the
machine, location and cause of damage. Based on the measurement history can reveal the remaining life
of the machine or until the damage becomes critical. The spectrum frequency also presents the amplitude
of the vibration in each source of damage occurring. Therefore, we can track where the location of the
damaged component is based on its frequency. Every damage to the machine creates a frequency that
has its own characteristics so that it can be used for diagnosis [27]. Vibration characteristics or vibration
signatures are obtained when the machine is operating. Actual vibration signals derived from transducers
can be considered as frequency characteristics, but usually refer to spectrum frequencies. Vibration
signal measurement with standard transducer assisted acceleration produces accurate vibration analysis.
The FFT is guided by the principle that all signals (not ideal signals) can be generated by the sum of
sine waves. Conversely any signal can be broken down on its components in the form of a sine wave, it
is important to say that the frequency spectrum fully represents the vibration signal. No information is
lost because of the conversion from time domain to spectrum frequency, if the phase difference between
components is also included [28, 29].
Spectrum characteristics provide valuable information about the damage location. The frequency and
magnitude in the spectrum are related to machine rotation. For example, an amplitude line appears on
1X, 2X, 3X, ... 7X or something else. The relationship that belongs to each frequency line, for example
the line that appears at 1X is greater than 2X or appears harmonic or there are sidebands. Knowing the
source of vibration amplitude originates, for example 1X is an unbalance fan, 5.34 is bearing damage,
etc.

2.3. Vibration assessment

Figure 1 shows type of ISO provides a standard for measurements carried out on site regarding the level
of vibration [30]. The standard set applies to machines that have a capacity more than 15 kW and
operating speeds in the middle of 120 RPM and 15000 RPM.

Figure 1. 10816-3 ISO Standard

4
ISAIME 2019 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 694 (2019) 012001 doi:10.1088/1757-899X/694/1/012001

3. Implementation
3.1. System analysis
Figure 2 shows the process of intelligent system. The input data in diagnosing the gearbox damage is in
the form of gearbox specification data including gearbox RPM input, gear RPM, number of teeth on
gear and pinion, number of ball bearings, vibration value and also input spectrum data. The data already
entered will be processed to find out the type of frequency that appears in the data spectrum.
Furthermore, the identification will be used as evidence to indicate whether or not there is a spectrum
line related to the damage to the gearbox. Identifying the frequency using predefined formulas. Suppose
the obtained frequency value of GMF is 275 Hz, then the system will see whether there is a line that
appears at that frequency or not. Existing evidence will be processed by calculating its probability value
using BNs method.
Based on the input data obtained will be calculated probability value through several stages using
BN method, which starts from the determination of parameter values for each symptoms of damage,
then determine the value of conditional probability, after the two values obtained then the system will
be calculated the value of joint and posterior probability of each type of gearbox damage that is adjusted
to the BNs structure and the posterior probability value is used as the probability inference of the type
of gearbox fault. BN produces relational information and conditional probabilities through two-way
propagation between input and output nodes. Also, in the application of real cases generally use
multistate nodes. So, from the consultation conducted by the user will get the type of fault that occurs
on the gearbox and the percentage of the fault. Users will know what action takes precedence.

Start

Gearbox Specification
Gearbox Type
RMS Value

No Correct

Yes

Spectrum Line
Identification

Yes

Evidence Bayesian Networks

Value Set No
2 Method Process

Yes

Gearbox Condition
Evidence Showing Inference
Information
1 Probabilistic

End No Reset

Figure 2. Intelligent system flowchart

5
ISAIME 2019 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 694 (2019) 012001 doi:10.1088/1757-899X/694/1/012001

3.2. Bayesian Networks approach

Designing expert systems in accordance with the concept map that you want to address in solving
problems. This concept supports the decomposition of knowledge that explains each component and
work procedure as discussed in figure 2, then created an expert system flowchart to illustrate in detail
the process flow of the system works.
According to the stage of the vibration analysis in gearbox fault detection, we can build the structure
of BN. In the first process, we assume the cause of the damage that occurs in the gearbox based on the
emerging spectrum line. Then built a BN in such a way as to be able to show what kind of damage and
what actions should performed by the user. The structure BNs model is shown in Fig. 3. Corresponding
to Pearl, relational probability distribution is obtained from expert knowledge [24]. Information on
object history is useful in determining prior probability values. This type of information is usually
available in every corporation for instance machine damage report. So here BN acts as a knowledge
base while the part of the control system is Bayesian inference rules.

S01 S02 S03 S04 S05 S06

S07 S08 S09 S10 S11 S12

S13 S14 S15 S16 S17 S18

HA HB LF GMF GFF BFr

Har MechL GM SM BFault GFault

VG CAS
Re Mo CSD

CAGS
Good IGF

CF

Bad A B
CB

C D
VB CGP

S19 S20 S21 S22 S23 S24

`
S25 S26 S27 S28 S29 S30 S31

Information:
Input Parameters Result
RMS Frequency
Action
Value Type Transition
Spectrum Nodes
Condition Conclusion
Line
Figure 3. Bayesian Networks system

6
ISAIME 2019 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 694 (2019) 012001 doi:10.1088/1757-899X/694/1/012001

There are 60 nodes in BNs structure, as shown in figure 3. In solving the problem of a large number
entering conditional probability value and to further ease the computer work, some nodes are added to
the network such as IGF, VG, Good, Bad, VB, HA, HB, etc. For instance, at node HA, there are five
parent nodes which are interconnected. The HA node will generate 64 CPT values, so time and memory
will be more needed if the root node has more parent nodes. Associated CPT held every root node, its
slump, only a row of representative prior probabilities. So, when a network has many parent nodes or
parent nodes containing a lot of classification values, CPT will also be very large. The CPT criterion
decided by the number of parent nodes. CPT for Boolean variables with x parent node having a 2x + 1
probability formula. Nodes such as Monitoring, Repair and Check Data Spectrum in Detail (CSD) are
added to serve to infer what actions should be performed by the user. Previous probabilities are needed
to make casual dependencies.
The CPT table shows the level of damage to the gears. An expert is required to fill each table with
a percentage of 0% - 100%. Suppose that the chance of occurrence of GMF if obtained evidence of
symptoms of S19 and S20 is 90%. Determination of the value of 90% is obtained from information
obtained by an expert. In analysing the damage of the gearbox using vibration data, the vibration value
generated on the velocity spectrum is very small because generally gear mesh frequency (GMF) occurs
at high frequency, and this will affect the small value of the resulting vibration. However, although the
value is small does not mean there is no damage to the gears. While one of the main advantages of the
BN when additional evidence is given, the BN system can determine conditional probability values in
each node and calculate more accurate probability propagation. In the diagnosis of gearbox damage,
conditional probabilities provide important information.

3.3. Software implementation

Material testing uses vibration data of multi stage gearbox, as shown in figure 4, that measurement was
taken from a synthetic yarn factory where is the data contains several kinds of measurement such as
overall velocity and spectrum velocity data.

Figure 4. Multi stage gearbox

gearbox has a rigid type foundation. There are 10 gears that rely on each shaft and bearings.
Measurement of vibration is taken at 4 points of shaft and each point is taken 3 axis point covering
horizontal, vertical and axial. Each of the 3 points is taken which has the greatest vibration value. Figure
5 shows the spectrum obtained from gearbox measurements at each accessible bearing location. There
are 4 spectrums chosen because it has the largest overall vibration value. Figure 6 describes the gearbox
structure in detail including the type of bearing and number of teeth that are used. According to the
complete specification information of the employee staff, the gearbox is grouped into ISO group 2 rigid
standard.

7
ISAIME 2019 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 694 (2019) 012001 doi:10.1088/1757-899X/694/1/012001

Figure 5. Spectrum of gearbox

8
ISAIME 2019 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 694 (2019) 012001 doi:10.1088/1757-899X/694/1/012001

Figure 6. Gearbox structure

3.4. Diagnosis result

Figure 7 shows the calculation results of BNs that have been built, there are symptoms of possible
indications of damage to the gears and bearings.

Figure 7. Diagnosis result of point 1

9
ISAIME 2019 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 694 (2019) 012001 doi:10.1088/1757-899X/694/1/012001

On the handling probability side, the highest percentage value is monitoring, this is because the vibration
value on the gearbox is still in good condition. As noted earlier, in particular, gears and bearing
problems, the small vibration values cannot be used as a benchmark to determine that the gears and
bearings are still in good condition or vice versa. Users should still check the state of the gears and
bearings according to experience or history of the gearbox.
Table 1. Fault detection and diagnosis result
Measurement Point 1A 1B 2 3 4
Gear 0.54 0.55 0.52 0.55 0.2
Probability

Gear Mis. 0.17 0.15 0.15 0.59 0.15

Fault

(%)

Foundation 0.42 0.42 0.42 0.42 0.42

Bearing 0.46 0.46 0.46 0.48 0.47
Shaft Mis. 0.28 0.28 0.28 0.28 0.29
Monitoring 0.49 0.49 0.49 0.49 0.5
Action
Prob.
(%)

Repairing 0.41 0.41 0.41 0.41 0.41

Checking Spectrum 0.25 0.25 0.25 0.25 0.25

Table 2. Comparison of Diagnosis result

No Diag. Point Symptoms at Spectrum Manual Diagnosis Intelligent System
Gear Mesh, Hunting Tooth, Gear Phase
0.3X-0.7X, 3.5X, Gear, GMF, Gear and Bearing
1 1A Assembly and Inner Race Bearing
GAPF, FHT, GFN and BPFI Problems
Problems
0.3X-0.7X, 3.5X, GMF, 2GMF, Gear Mesh, Hunting Tooth, Gear Phase
Gear and Bearing
2 1B 5GMF, GAPF, 2GAPF, FHT, Assembly, Outer Race Bearing and
Problems
GFN, BPFO and 2BSF Ball Bearing Problems
0.3X - 0.7X, 5GMF, FHT, BPFI Gear Mesh, Hunting Tooth, Inner Race Gear and Bearing
3 2
and FTF Bearing and Cage Bearing Problems Problems
Gear Misalignment, Gear Mesh,
0.3X-0.7X, Gear, 2Gear,
Hunting Tooth, Gear Phase Assembly,
Pinion, 2Pinion, GMF, 5GMF, Gear, Mis. Gear and
4 3 Inner Race Bearing, Outer Race
GAPF, 2GAPF, FHT, GFN, Bearing Problems
Bearing, Ball Bearing and Cage
BPFI, BPFO, 2BSF and FTF
Bearing Problems
1X, 0.3X - 0.7X, BPFO, 2BSF Outer Race Bearing, Ball Bearing and
5 4 Bearing Problems
and FTF Cage Bearing Problems

According to a series of diagnostic tests that have been done on the gearbox, we can know easily
what damage happens to the gearbox. The results of each calculation using different BNs methods in
each trial, this is due to the difference in evidence obtained from each spectrum data against the gearbox
specification data. We can clearly examine the evidence data that is entered on the BNs method. Based
on evidence of symptoms obtained, we can conclude whether the results of the diagnosis are appropriate
or not. The results of each diagnosis indicate the suitability of the design with the intended expectations.

4. Conclusion
Designing an intelligent system, which considers both vibration analysis and BN system with multiply
states, using discrete failure probability distribution. BN based on the vibration frequency characteristic
analysis procedure applied in the implementation system to better present BN applications in expert
systems so as to produce accurate information from the gearbox condition. The advantages of BN are
indicated by two-way inference for conditional probabilities. Damage to the components of the gearbox
can be seen after proof of the spectrum added to the BN. A more accurate and faster decision can be
made to ensure the performance of the expert system and provide valuable information regarding
damage detection and gearbox maintenance. Results of diagnostic testing obtained matching results
between the tests conducted manually and by the expert system. BNs method have been effectively used
to the diagnostic expert system for the source of gearbox damage, so it can provide a fast diagnosis and
the probability of occurrence of the location of the damage.

10
ISAIME 2019 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 694 (2019) 012001 doi:10.1088/1757-899X/694/1/012001

5. Acknowledgments
This work is financially supported by the Research Project of Mechanical Engineering in Beijing
Institute of Technology and Mercu Buana University. The research data supported by Putranata Adi
Mandiri, Inc.

References
[1] G. Diwakar, M. Satyanarayana and P. Kavi Kumar, "Detection of Gear fault using vibration
analysis," International Journal of Emerging Technology and Advanced Engineering, vol. 2,
no. 9, pp. 122-125, 2012.
[2] L. Jing, M. Zhao, P. Li and X. Xu, "A convolutional neural network based feature learning and
fault diagnosis method for the condition monitoring of gearbox," Measurement, vol. 111, pp.
1-10, 2017.
[3] A. Aherwar and M. S. Khalid, "Vibration Analysis Techniques for Gearbox Diagnostic: A
Review," International Journal of Advanced Engineering Technology, vol. 3, no. 2, pp. 1-7,
2012.
[4] J. Lee, F. Wu, W. Zhao, M. Ghaffari, L. Liao and D. Siegel, "Prognostics and health management
design for rotary machinery systems—Reviews, methodology and applications," Mechanical
Systems and Signal Processing, vol. 42, no. 1-2, pp. 314-334, 2014.
[5] M. Y. Asr, M. M. Ettefagh, R. Hassannejad and S. N. Razavi, "Diagnosis of combined faults in
Rotary Machinery by Non-Naive Bayesian approach," Mechanical Systems and Signal
Processing, vol. 85, pp. 56-70, 2017.
[6] D. N. Dongiovanni and T. Iesmantas, "Failure rate modeling using fault tree analysis and
Bayesian network: DEMO pulsed operation turbine study case," Fusion Engineering and
Design, vol. 109, pp. 613-617, 2016.
[7] W.-S. Wu, C.-F. Yang, J.-C. Chang, P.-A. Château and Y.-C. Chang, "Risk assessment by
integrating interpretive structural modeling and Bayesian network, case of offshore pipeline
project," Reliability Engineering & System Safety, vol. 142, pp. 515-524, 2015.
[8] X. Li, G. Chen and H. Zhui, "Quantitative risk analysis on leakage failure of submarine oil and
gas pipelines using Bayesian network," Process Safety and Environmental Protection, vol.
103, pp. 163-173, 2016.
[9] W. Wang, K. Shen, B. Wang, C. Dong, F. Khan and Q. Wang, "Failure probability analysis of
the urban buried gas pipelines using Bayesian networks," Process Safety and Environmental
Protection, vol. III, pp. 678-686, 2017.
[10] H. Langseth and L. Portinale, "Bayesian networks in reliability," Reliability Engineering &
System Safety, vol. 92, no. 1, pp. 92-108, 2007.
[11] I. Boersch, J. Heinsohn and R. Socher, Wissensverarbeitung: Eine Einführung in die künstliche
Intelligenz für Informatiker und Ingenieure, vol. 2, Elsevier, 2007, p. 397.
[12] K. A. Kobbacy, K. McNaught and A. Chan, "Bayesian networks in manufacturing," Jnl of Manu
Tech Mnagmnt, vol. 22, no. 6, pp. 734-747, 2011.
[13] J. Pearl, Bayesian Networks: a model of self-activated: memory for evidential reasoning, Los
Angeles: UCLA, Computer Science Department, 1985.
[14] J. Pearl, "Fusion, propagation, and structuring in Belief Networks," Artificial Intelligence, vol.
29, pp. 241-288, 1986.
[15] L. Yang and J. Lee, "Bayesian Belief Network-based approach for diagnostics and prognostics of
semiconductor manufacturing systems," Robotics and Computer-Integrated Manufacturing,
vol. 28, no. 1, pp. 66-74, 2012.
[16] Z. Xuejun and C. Yu, "The research of intelligent fault diagnosing methods based on FTA,"
Microcomputer Information, vol. 21, no. 6, pp. 123-124, 2005.
[17] Y. Zhi-Ling, W. Bin, D. Xing-Hui and L. Hao, "Expert System of Fault Diagnosis for Gear Box
in Wind Turbine," The 2nd International Conference on Complexity Science & Information
Engineering, vol. 4, pp. 189-195, 2011.

11
ISAIME 2019 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 694 (2019) 012001 doi:10.1088/1757-899X/694/1/012001

[18] N. Khakzad, F. Khan and P. Amyotte, "Safety analysis in process facilities: comparison of fault
tree and Bayesian network approaches," Reliability Engineering & System Safety, vol. 96, no.
8, pp. 925-932, 2011.
[19] N. Khakzad, F. Khan and P. Amyotte, "Dynamic safety analysis of process systems by mapping
bow-tie into Bayesian network," Process Safety and Environmental Protection, vol. 91, no. 1-
2, pp. 46-53, 2013.
[20] N. Khakzad, F. Khan and P. Amyotte, "Quantitative risk analysis of offshore drilling operations:
a Bayesian approach," Safety Science, vol. 57, pp. 108-117, 2013.
[21] B. G. Xu, "Intelligent fault inference for rotating flexible rotors using Bayesian belief network,"
Expert Systems with Applications, vol. 39, pp. 816-822, 2012.
[22] D. Zhou, “The Application of Bayesian Networks in System Reliability,” Master Thesis in
Arizona State University, pp. 1-13, 2014.
[23] J. Grover, Strategic economic decision-making: Using Bayesian belief networks to solve complex
problems, vol. 9, Springer Science & Business Media, 2012.
[24] J. Pearl, Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference (Morgan
Kaufmann Series in Representation and Reasoning), San Mateo, California, 1988.
[25] K. B. Korb and A. E. Nicholson, Bayesian Artificial Intelligence (Second Edition), London: CRC
Press, 2011.
[26] R. E. Neapolitan, Learning Bayesian Networks, Chicago, Illinois: Northeastern Illinois
University, 2003.
[27] J. Mais, "Spectrum Analysis: The key features of analyzing spectra," SKF USA Inc., pp. 3-28,
2002.
[28] S. K. Sar and R. Kumar, "Techniques of Vibration Signature Analysis," International Journal of
Advanced Research in Computer and Communication Engineering, vol. 4, no. 3, pp. 240-243,
2015.
[29] S. R. W. Mills, Vibration Monitoring and Analysis Handbook, vol. 1, Northampton: The British
Institute of Non-Destructive Testing, 2010.
[30] NEN, "Mechanical Vibration - Evaluation of machine vibration by measurements on non-rotating
parts," International Standard, vol. II, pp. 1-10, 2009.

12