We are IntechOpen,

the world’s leading publisher of

Open Access books
Built by scientists, for scientists

3,350
Open access books available
108,000
International authors and editors
1.7 M
Downloads

Our authors are among the

151
Countries delivered to
TOP 1%
most cited scientists
12.2%
Contributors from top 500 universities

Selection of our books indexed in the Book Citation Index

in Web of Science™ Core Collection (BKCI)

Interested in publishing with us?

Contact [email protected]
Numbers displayed above are based on latest data collected.
For more information visit www.intechopen.com
 Chapter 14

Artiicial
Artificial Intelligence
Intelligence Application
Application in
in Machine
Machine Condition
Condition
MonitoringMonitoring and Fault Diagnosis
and Fault Diagnosis

Yasir Hassan Ali

Additional information is
Additional information is available
available at
at the
the end
end of
of the
the chapter
chapter

http://dx.doi.org/10.5772/intechopen.74932

Abstract
The subject of machine condition monitoring and fault diagnosis as a part of system
maintenance has gained a lot of interest due to the potential beneits to be learned from
reduced maintenance budgets, enhanced productivity and improved machine availabil-
ity. Artiicial intelligence (AI) is a successful method of machine condition monitoring
and fault diagnosis since these techniques are used as tools for routine maintenance. This
chapter atempts to summarize and review the recent research and developments in the
ield of signal analysis through artiicial intelligence in machine condition monitoring
and fault diagnosis. Intelligent systems such as artiicial neural network (ANN), fuzzy
logic system (FLS), genetic algorithms (GA) and support vector machine (SVM) have pre-
viously developed many diferent methods. However, the use of acoustic emission (AE)
signal analysis and AI techniques for machine condition monitoring and fault diagnosis
is still rare. In the future, the applications of AI in machine condition monitoring and fault
diagnosis still need more encouragement and atention due to the gap in the literature.

Keywords: artiicial intelligence, machine condition monitoring, fault diagnosis

1. Introduction

In the current commercial production industries, there is an increasing trend towards the
need for higher availability equipment that can work nonstop 24/7. Thus, any type of fail-
ure, even minor, cannot be accepted as it can signiicantly afect the cost and the production.
Hence, a very accurate monitoring of the machine condition and a proper fault diagnosis of
the machine failure is necessary. The machine fault diagnosis had seen a vast improvement
since the maintenance was provided after the machine had developed a fault and afected the

© 2016 The Author(s). Licensee InTech. This chapter is distributed under the terms of the Creative Commons
© 2018 The Author(s). Licensee IntechOpen. This chapter is distributed under the terms of the Creative
Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use,
Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use,
distribution, and reproduction in any medium, provided the original work is properly cited.
distribution, and reproduction in any medium, provided the original work is properly cited.
276 Artificial Intelligence - Emerging Trends and Applications

production. After that, it developed into preventive maintenance in the past few years before
all the industries started using the condition-based maintenance. Preventive maintenance can
be deined as providing maintenance before the machinery faces any fault.

On the other hand, condition-based maintenance can be deined as providing maintenance

depending on the data obtained from target measurements. The eiciency of this technique
is measured depending on the accurate diagnostic tactics, which are fulilled. For surviving
in the current competitive market, the industries need to improve their product reliability
and also reduce their production costs. The product reliability is more important for speciic
productions like for the aviation, nuclear and the petrochemical industries where any failure
can lead to severe environmental disasters.

Currently, industries have shifted from using the condition-based (predictive) approach to
the maintenance-based approach depending on the trending and the data analysis from one
or more parameters that indicate the development or the presence of known failures or faults.
The efective machine condition monitoring technique must be able to determine the onset of
any fault in its early stages and also provide an accurate diagnosis regarding the type of the
fault and its location. Ideally, the condition monitoring technique must give an overall and a
detailed accurate health assessment of the equipment.

However, conventionally, it would include the aural and the visual inspection (applying all
the human senses), temperature monitoring, oil analysis (known as the wear debris analysis),
measurement of the vibrations and its analysis, motor current signature analysis, airborne
sounds and the acoustic emission (AE) analysis. In acoustic emission analysis, the waves are
sent from an emission source and transferred to the surface by the transmission medium. The
low-displacement or high-frequency mechanical waves can be picked up as electronic signals.
The signal strength can be increased by using a preampliier before the data are interpreted
by the AE equipment [1, 2].

Furthermore, there is a growing interest in developing new technologies to overcome the

problems in condition monitoring and diagnostics of complex industrial machinery appli-
cations, which were not resolved till now. This provides excellent opportunities for the AI
technology to grow continuously, with the rapid increase in the growth of intelligent informa-
tion, sensor and data acquisition capabilities, combined with the rapid advances in intelligent
signal processing techniques [3]. AI techniques that have been extensively used in the ield
of engineering include genetic algorithms (GA), support vector machine (SVM), fuzzy logic
system (FLS) and artiicial neural network (ANN). As compared to the common fault diag-
nostic approaches, the AI techniques are instrumental if they can be improved [4]. Apart from
improving performance, these techniques can be easily extended and modiied. These can be
made adaptive by the integrating new data or information [5].

In this chapter, an atempt has been made to review the recent developments in the ield of
acoustic emission signal analysis for fault diagnostics of the machine based on the aforemen-
tioned AI techniques. These systems can be mutually integrated with each other and also with
other traditional techniques.
 Artificial Intelligence Application in Machine Condition Monitoring and Fault Diagnosis 277
http://dx.doi.org/10.5772/intechopen.74932

2. Artiicial intelligence

The AI is the system that thinks and acts like a human being. It can also imitate human
behaviour. It is majorly concerned with the development of a computer’s ability to engage
in human-like thought processes like learning, reasoning and self-correction [6]. In the last
decade, there has been a growing need in AI to solve the problems of engineering. Earlier,
these problems were considered hard to be solved analytically or by using mathematical
modelling and needed human intelligence [7].

Nowadays, there is an increased demand for advanced AE analysis tools. This chapter shows
that many scholars have studied the detection and diagnostic of several faults by using the AE
methods in AET and signal analysis. The AI techniques as mentioned earlier have also been
extensively used in the ield of engineering.

2.1. Artiicial neural network-based fault diagnosis

Artiicial neural network (ANN) is an information-processing approach. It works like the

biological nervous systems like how the brain processes the information in the human body.
The discussion was limited to an introduction of many components, which were involved in
the ANN implementation. The network architecture or topology (including number of nodes
in hidden layers, network connections, initial weight assignments and activation functions)
played a key role in the ANN performance and depended on the problem at hand. Figure 1
shows a simple ANN and its constituents. In most cases, seting the correct topology was based
on a heuristic model. On the other hand, the dimensions of the input and the output spaces gen-
erally suggested the number of input and output layer nodes. Selecting the network complexity
or regularization was very important [8]. When designing a neural network, there are a number
of diferent parameters that must be decided. Some of these parameters are the number of

Figure 1. Architecture of a neural network.

278 Artificial Intelligence - Emerging Trends and Applications

training iterations, the number of layers, the learning rate, the number of neurons per layer
and the transfer functions, and so on.

The beneit of ANN was that it had the ability to respond to an input patern in a desirable
manner after the learning phase. Previous studies have proved that the eiciency of ANN can
predict the faults of machining processes. This technique was found to be very useful as it can
be used in industrial automation in a more lexible manner [9]. ANN has been extensively
used in health diagnosis of mechanical gear, bearing and rotating machines by using features
more from vibration signals and less from the acoustic signals.

There has been an increasing demand for advanced AE analysis tools, which have the capacity to
distinguish diferent sources of AE data. This has resulted in developing modern and more lex-
ible patern recognition software, combining traditional, graphical AE analysis and advanced
unsupervised patern recognition (UPR) and supervised patern recognition (SPR) analysis.
Application of the UPR techniques on AE data during various test cases has also increased the
understanding of the damage evolution and the capacity of noise discrimination [10].

The problem of a roller with health monitoring has illustrated the efectiveness of GA for fault
classiication by using ANNs [8]. In this regard, Al–Balushi and Samanta have suggested a
procedure to diagnose the fault of gears by wavelet transformation and ANN for AE signals.
These features were taken from wavelet transformation and were used as an input to an ANN
based on diagnosing approach [11]. In the fault prognosis systems, the acoustic emission and
vibration signal were utilised as an input signal. Additionally, ANN was utilised as a prog-
nosis system for rotating machinery failure [12]. In this way, a multiple-layer neural network
was successfully used to detect the fault in the gearbox, and classiication was used to utilise
the supervised learning with an experimentally obtained data. The data were presented as
processed vibration and acoustic emission signals [13]. The utilisation of acoustic emission
for early detection of the helicopter rotor head dynamic component faults was previously
studied. They analysed the stress wave of the light-test data set by using the wavelet-based
techniques for assessing the background operational noise as compared to machinery failure
results. The feed-forward neural network was used as a classiier to determine the correct
ight regime [14].

For solving the issues of velocity and the time diferences, a new approach to AE source
localization was described. This new approach to AE source location was documented on
the wing spar cut-out of L-39 aircraft, as this method was used to estimate the AE source
coordination by using the ANN process which extracted signal parameters [15]. Fog et al.
studied the detection of the exhaust valve burn—through a four-cylinder, 500 mm bore and
two-stroke marine diesel engine. This investigation comprised of monitoring three diferent
valve conditions (normal, leak and large leak). Vibration and structure-born stress waves (AE)
were monitored. The acoustic emission (AE) signal features were extracted by using principal
component analysis (PCA). A feed-forward neural classiier was also used for discriminating
between the three valve conditions [16].
The AE data collected during a static test of a 12-m FRP wind turbine blade was analysed and
classiied by using diferent unsupervised patern recognition (UPR) techniques, and using
 Artificial Intelligence Application in Machine Condition Monitoring and Fault Diagnosis 279
http://dx.doi.org/10.5772/intechopen.74932

the UPR results, a supervised patern recognition (SPR) method was trained based on the
back-propagation neural network. This was applied to the AE data collection and a subse-
quent biaxial fatigue loading of the same blade [17].
The neural network gained atention in grinding research due to its functions of learning,
interpolation and patern recognition and classiication. Diferent other examples of the appli-
cation in the engineering ield were also reported [18–22]. Aguiar et al. atempted to atain the
classiication of burn degrees of the surface-grinding machine, which was utilised for grind-
ing tests with an aluminium oxide-grinding wheel and the utilisation of neural networks.
The AE and power signal along with the statistics from the digital signal processing of these
signals were used as inputs of the neural networks [9].

Furthermore, the ANN approach was proposed for the detection of work-piece “burn”, the
unwanted change in metallurgical properties of the material produced by overly aggressive
or otherwise inappropriate grinding [19]. The grinding AE signals for 52100 bearing steel
were collected and digested to extract feature vectors. These appeared to be more useful for
ANN processing. Aguiar et al.’s work was diferent as it used grinding parameters as an input
to the neural networks that were not tested yet in surface roughness prediction by neural
networks. In addition, a higher sampling rate data acquisition system was used to get the
acoustic emission and cuting power [23].

Goebel and Wright developed hybrid architecture, featuring fuzzy logic and neural net-
works to cope with weaknesses of traditional methods for monitoring and diagnosing an
unatended milling machine. Force, spindle current and acoustic emission data were used
as inputs to the neural network after they underwent some signal processing for calculating
the membership functions of fuzzy relations. Additionally, fuzzy logic principles were uti-
lised for diagnosing the system's status concerning tool wear and chater [24]. The indings
of it was encouraging to use neural network in detection and classiication of work piece
“burn” and surface roughness prediction revealed that AE signal from grinding machine
[9, 19, 23, 24].
Impact damage is a problem that damages the composite industry. This damage may seem
supericial, but it may often have negative efects on the performance of the composite struc-
ture. The conventional NDE techniques can detect the locations or the shapes of the impact
damage and cannot quantify its efects on the structure. Conversely, AE records the active
law growth when the structure gets loaded. It also measures the reduction in the structural
performance produced by an impact load. AE signal analysis was used to measure the efect
of impact damage on burst pressure in 5.75 inch diameter, inert propellant-illed and ila-
ment-wound pressure vessels. The AE data were collected from 15 graphite/epoxy pressure
vessels featuring 5 damage states and 3 resin systems. A burst pressure prediction model was
developed by correlating the AE amplitude (frequency) distribution, generated during the
irst pressure ramp, to 800 psig to known burst pressures using a four-layered back-propaga-
tion neural network [25].
The ANN patern recognition technique was used for analysing the AE source signals of
the pressure vessel in the site. For this purpose, a new quantitative analysis concept for AE
280 Artificial Intelligence - Emerging Trends and Applications

sources of pressure vessel was introduced by using artiicial neural network classiication
along with raising a new method to evaluate the severity of the AE sources [26].

Macías conducted an analysis of the relationship between AE signals and the main parameters
of friction stir welding (FSW) process on the basis of ANN. The AE signals were acquired by
data acquisition, applied in the welding process, carrying out plates 3 mm thick of aluminium
alloy. Wavelet transform (WT) was also used for the statistical and temporal parameters of the
decomposition of EA signals as input for the multilayer feed-forward ANN [27].

The partial discharge (PD) detection, signal analysis and patern identiication, using AE mea-
surements and the back-propagation (BP) ANN, were also studied. In this way, the measured
signals were processed with three-dimensional paterns and short duration Fourier trans-
forms (SDFT). The indings showed that utilisation of BP ANN with the SDFT components for
the classiication of the diferent PD paterns provided excellent results [28].
To determine the quality of feature extraction and for the ANN classiier, performances were
also conducted through a series of experimentations. This helped in input data acquisition
during AE experiments on the chemical process plant. These input data consisted of a set of
AE power spectra. Each source input data ile was subjected to preprocessing consisting of
additional linear averaging in each input vector and individual amplitude normalisation by
removing the mean value and division by the standard deviation of the feature. Three-layer
networks using the back-propagation updating scheme were used for assessing their com-
bined feature extraction and classiication capabilities, while solving the problem of process
stage recognition [29].

Until 2015, there is no study in the literature related to the estimation of oil ilm thickness
through acoustic emission signals, so to predict and monitor oil ilm thickness of spur gear, a
test rig was built and the gearbox was run at diferent speeds and load conditions. Artiicial
neural network (ANN) and regression models used to predict the lubricant regime depended
on oil temperature, acoustic emission signals and speciic ilm thickness (�). Both FFBP and
Elman network models were used to predict speciic oil ilm thickness with input as AE and
temperature data. The results showed that FFBP and Elman models were efective in predict-
ing oil ilm thickness from acoustic emission signal and temperature, and this suggested tech-
nique atained 99.9% success in prediction and classiication at high speed during training.
The FFBP was beter than Elman during testing and gave excellent results in prediction and
classiication. Thus, the architecture and topology of the network through speciic systems
can be used for online monitoring of oil ilm thickness and to predict any causes of failure of
spur gear operation [30, 31].

2.2. Spiking neural network

Recently, spiking neural network (SNN) is the third-generation neural network (Figure 2)
and has gained a lot of interest in the scientiic community [32]. The SNNs became famous
before the introduction of the sigmoidal or the perceptron neuron [32]. It was observed that
the SNNs were very suitable for the parallel implementation in the digital hardware [32] and
in the analogue hardware [33, 34].
 Artificial Intelligence Application in Machine Condition Monitoring and Fault Diagnosis 281
http://dx.doi.org/10.5772/intechopen.74932

Figure 2. Spiking neural network [32].

The earlier generations of the neural networks used the analogue signals for conveying the
data from one neuron to the next. This communication between the neurons in the SNNs
used spikes, which was similar to the system used in the actual human neurons. The spikes
could be recognized only at those instances when they had occurred. With the help of the
weighted sum of the analogue input value, the earlier neuron estimated the value using the
sum-speciic non-linear function. The value helped in determining the delay in the spike out-
put, which was aimed for the succeeding neuron. Generally, the spiking neuron was viewed
as the leaky integrator because the target neuron integrated the spikes for a period of time
and accepted the resultant integrated values used as the membrane potential. When the mem-
brane potential value approached a speciic threshold value, then, the neuron was seen to
send a spike; thereafter, the membrane potential value was reset.
An increased knowledge in the information processing of the biological neurons helped in
explaining many additional parameters (like the gene and the protein expression) that needed
to be taken into consideration for the neurons to spike [33–35]. The additional parameters
included the diferent physical properties of the connections [32], the likelihood of the spikes
being accepted at the synapse and the emited neurotransmiters or the open-ion channels
[36, 37]. Several of the properties were modelled mathematically and were used for studying
the biological neuronal system [38, 39]. The SNNs were made of the artiicial neurons that
communicated using the trains that were considered as the pulse-coded data [40]. The SNN
was biologically acceptable, and it was seen to ofer a means for the representation of the
frequency, time, phase and such other features for the information processing. Moreover, the
SNN possessed the ability for training the neurons for converting their spatial-temporal data
to spikes (their properties include the spiking rates and spiking time). When one was select-
ing the neuronal model for an SNN, one needed to consider the computational eicacy and
the biological credibility [40]. If it was seen that the computational eicacy was beter than
the biological plausibility, then the leaky integrate-and-ire (LIF) model needed to be adopted
due to its cost efectiveness.
282 Artificial Intelligence - Emerging Trends and Applications

In their study, Silva et al. [41, 42] depicted the applications of the prototype decision support
system for monitoring the tool wear depending on the SNN technique. This system consisted
of six diferent components, that is, collection of data, feature extraction, multi-sensor inte-
gration, patern recognition, tool wear estimation and the outlier detection. Their proposed
architecture consisted of one built-in self-organizing neural architecture part that was based
on the SNN. Their study showed that the modelling process was very eicient for classify-
ing the tool wear level of the tool inserts with the help of the apparent weak features. Their
method showed the efectiveness of using the SNN model for the tool condition monitoring,
thus implying that the approach was feasible for many industrial applications, wherein a
lot of noisy data are obtained. This researcher was the only one who used SNN in condition
monitoring; the result showed the capability of spiking neuron networks for tool condition
monitoring.

2.3. Genetic algorithm-based fault diagnosis

GA created by John Holland in the 1970s is an evolutionary algorithm which is part of the ield
of artiicial intelligence. A genetic algorithm (GA) is a method for solving both constrained
and unconstrained optimization problems based on a natural selection process that mimics
biological evolution. The algorithm repeatedly modiies a population of individual solutions.
At each step, the genetic algorithm randomly selects individuals from the current population
and uses them as parents to produce the children for the next generation. Over successive
generations, the population “evolves” towards an optimal solution.

As originally proposed, a simple GA mainly consists of three processes: selection, genetic

operation and replacement. Description of a typical GA cycle and its high-level description
are provided in Figure 3. The population composed of a group of chromosomes, which were
the candidates for the solution. The itness values of all chromosomes were evaluated by an

Figure 3. Genetic algorithm cycle.

 Artificial Intelligence Application in Machine Condition Monitoring and Fault Diagnosis 283
http://dx.doi.org/10.5772/intechopen.74932

objective function (performance criteria or a system’s behaviour) in a decoded form (pheno-

type). A particular group of parents was selected from the population for generating ofspring
on the basis of the deined genetic operations of crossover and mutation.

The itness of all ofsprings was then evaluated using the same criterion. The chromosomes in
the current population were then replaced by their ofspring on the basis of a certain replace-
ment strategy. Such a GA cycle was repeated until the termination criterion was reached [8].
Using ANNs utilised a simple problem of a roller with health monitoring to illustrate the
efectiveness of GA in AE feature selection for fault classiication. It revealed that utilising GAs
to select an optimal feature set for a classiication application of ANNs was a very powerful
technique [8]. Ming applies the AE technique for bearing condition monitoring and fault diag-
nosis. Scales for continuous wavelet transform, wavelet-based waveform parameter selection
and optimisation on the basis of genetic algorithm were the proposed selection methods [43].

The AE was monitored by utilising a data acquisition system during the process of conduct-
ing the mechanical tests on several materials. Two of the sensors were positioned directly
on the specimen. AE signals were thought to be patern vectors described by a number of
writers. In this chapter, “model” data sets were generated to become closer to AE signals that
were recorded during the tests. This chapter presented and validated a genetic algorithm-
based approach to cluster the AE signals. Its superiority over the k-means algorithm was
highlighted by the study of diferent “model” data sets. The genetic strategy can be character-
ised by a high stability and a high performance especially to cluster data sets consisting of a
minority class, a cluster with signals of extreme features or a set of clusters with very diferent
sizes [44].

2.4. Fuzzy logic-based fault diagnosis

Zadeh introduced the fuzzy logic (FL) in 1965 [45–47]. FL is a multi-valued logic that allows
the intermediate values between conventional evaluations like true/false, yes/no, high/low,
and so on. The FL helps in providing a variety of ways to solve a control or classiication
problem. Thus, this method focuses on what the system should do rather than trying to model
how it works [48].

Aguiar et al. work is mentioned twice in this chapter because it contains two parts: irst part
used ANN for the classiication of burn degrees of the surface grinding machine; in this part,
a methodology was used to predict the surface roughness of advanced ceramics by using an
adaptive neuro-fuzzy inference system (ANFIS). For this study, alumina work pieces were
pressed and sintered into rectangular bars. The statistical data processed from the AE signal
and the cuting power were also used as input data for ANFIS [9]. Cusido et al. provided
approaches for a one-board fault detecting system and test program set (TPS) fault detecting
system for electromechanical actuators (EMA) ball bearings by analysing the diferent vibra-
tion and AE signals and by using FL inference techniques [49].
Omkar et al. presented the results of fuzzy modelling to discover the problem in grinding
through digital processing of the acoustic emission signals produced during the process.
Fuzzy C-means (FCM) clustering was utilised in classiication of the AE signal to diferent
284 Artificial Intelligence - Emerging Trends and Applications

sources of signals. FCM was potentially helpful in discovering the cluster among the data,
when the boundaries between the subgroups overlap. AE test was conducted by using pulse,
pencil and spark signal source on the surface of solid steel block. Four parameters such as
event duration, peak amplitude, rise time and ring down count were measured with the help
of AET 5000 system. These data were then used in training and validation of the FCM-based
classiication [50].

Aguiar et al. investigated the burning in the grinding process on the basis of a fuzzy model.
The inputs of the models were received from the digital processing of the raw AE and cuting
power signals. The parameters obtained and used in this study consisted of the mean-value
deviance, grinding power and root mean square (RMS) of the acoustic emission signal [51].
Ren et al. also atempted to introduce the most successful AE model during the continuous
cuting periods by using fuzzy modelling. The fuzzy identiication method provided a simple
way to arrive at a more deinite conclusion on the basis of the information collected with the
diiculty in understanding the exact physics of the machining process [52].

Recent studies used type-2 fuzzy logic in their research [53–56] because of the need to have
extremely fuzzy situations to use type-2 fuzzy. If we were extending the use of FL to a higher
order, then it is called type-2 FL. Hence, Ren et al. explained how type-2 TSK [Takagi–Sugeno–
Kang (TSK)] fuzzy uncertainty estimation method was implemented to ilter the raw AE sig-
nal directly from the AE sensor during turning process. This paper speciically focuses on
iltering and capturing the uncertainty by type-2 TSK fuzzy approach on the interval of AE
signal during one 10 mm cuting length [53].

Ren et al. atempted to ind out the relationship between AE and tool wear. They presented
an application of type-2 FL on AE signal modelling in precision manufacturing. Type-2 fuzzy
modelling was used for distinguishing the AE signal in precision machining. It provided a
simple way for arriving at a deinite conclusion without understanding the exact physics of
the machining process [54].

The knowledge about uncertainty prediction of tool life was highly essential for tool condition
investigation. It was also important for taking decisions about how to maintain the machine
quality. Ren et al. presented a type-2 fuzzy tool condition monitoring (TCM) system based on
AE in micro-milling. In the system, type-2 FLSs were utilised for analysing the AE signal fea-
ture (SF) and choosing the most reliable ones for integration to efectively estimate the cuting
tool condition through its life. The acquired results show that the type-2 fuzzy tool life estima-
tion is in accordance with the cuting tool wear state during the micro-milling process [55].

A type-2 fuzzy analysis method was utilised to analyse the AE SFs in TCM in micro-milling
process. The interval output of type-2 approach provided an interval of uncertainty associ-
ated with SFs of AE signal. The SFs with less RMSE and variation were selected to estimate the
cuting tool life in the future [56]. The new philosophy for AE source localisation under high
background noise was also designed. The algorithm was based on probabilistic and fuzzy-
neuro principles, so AE events can be put to classiication according to their energy and loca-
tion probability. AE signals recorded during the stamping processes of a thin metal sheet
were used for new algorithm testing [57].
 Artificial Intelligence Application in Machine Condition Monitoring and Fault Diagnosis 285
http://dx.doi.org/10.5772/intechopen.74932

Khalifa and Komarizadeh developed an eicient walnut recognition system through put-
ting together the AE analysis, principle component analysis (PCA) and adaptive neuro-fuzzy
inference system (ANFIS) classiier. This new system was tested later and classiied walnuts
into two classes. In the classiication phase, selected statistical features were used as the input
for the ANFIS classiier [58].

2.5. Support vector machine-based fault diagnosis

The support vector machine (SVM) approach was utilised in the form of a classiication tech-
nique on the basis of the statistical learning theory (SLT). It was basically based on the principle
of hyperplane classiier or linear separability. The main purpose of SVM was to explore a linear
optimal hyperplane for maximizing the margin of separation between the two classes [59, 60].
The SVM was utilised for fault diagnosis of spur bevel gear box. This was considered to be
a popular machine learning application due to its higher accuracy and for its generalization
capabilities [61]. These studies also examined the fault diagnosis of low-speed bearings based
on AE technique and vibration signal. Fault diagnosis was conducted by using the classiica-
tion technique with the help of relevance vector machine (RVM) and SVM. The classiication
process provided a comparative study between RVM and SVM for fault diagnosis of low-
speed bearing [62, 63].
Yu and Zhou exposed the method to classify the AE signals in composite laminates by utilis-
ing SVM. The classiier had built to achieve the identiication and classiication of AE signals.
The results of simulation showed that SVM had the potential to efectively distinguish difer-
ent acoustic emission signals and noise signals. The classiication accuracy rate of grid search
parameters was higher than the GA algorithm by this method [64]. Chu-Shu also revealed
the method on how to classify the AE signals in composite laminates by using the SVM [65].

On the basis of a thorough review of literature, this study informs about the new approaches
on the basis of hierarchical clustering and support vector machines (SVM) and are introduced
to cluster AE signals and to detect P-waves for micro-crack location in the presence of noise
through inducing the cracks in rock specimens during a surface instability test [66]. Thus, this
chapter proposes a novel grinding wheel wear monitoring system based on discrete wavelet
decomposition and SVM. The grinding signals were collected by an AE sensor [67].

Elforjani used a model to analyse the output signals of a machine while in operation and
accordingly helps to set an early alarm tool that reduces the untimely replacement of com-
ponents and the wasteful machine downtime. In this work, Elforjani uses three supervised
machine learning techniques such as Gaussian process regression (GPR), support vector
machine regression (SVMR) and multi-layer artiicial neural network (ANN) model to corre-
late AE features with corresponding natural wear of slow-speed bearings throughout the series
of laboratory experiments. Analysis of signal parameters such as root mean square (RMS) and
signal intensity estimator (SIE) was done to discriminate the individual types of early damage.
It was concluded that neural network models with back-propagation learning algorithm have
an advantage over the other models in estimating the remaining useful life (RUL) for slow-
speed bearings if the proper network structure is chosen and suicient data are provided [68].
286 Artificial Intelligence - Emerging Trends and Applications

The development of AI technique shows a promising potential in machine condition monitor-

ing and diagnosis, although only few articles were found in this area. However, ANN based
on AE has been successfully applied to many relevant problems. It can be considered that
ANN is the most new popular method in condition monitoring with AE signal. The use of
fuzzy, GA and SVM in condition monitoring and fault diagnosis based on AE signal analysis
still needs additional atention because of the absence of available papers. Finally, the future
works will be able to ind a novel idea for machine condition monitoring and fault diagnosis
using AE signal analysis and AI [69].

3. Conclusion

This chapter presents a survey based on a literature review using AE signal analysis and AI
techniques in machine condition monitoring and fault diagnosis. It surveys the articles with
a keyword index machine condition monitoring and machine fault diagnosis using AE signal
analysis and AI.

We can conclude that the classiication of AE signals carries high importance in machine
condition monitoring and fault diagnosis. AI has several advantages when compared to the
traditional mathematical modelling and statistical analysis. This includes dispensing of the
necessity for detailed system behaviour knowledge, which can be replaced by relatively sim-
ple computational methods. ANN based on AE has been successfully applied to numerous
relevant problems. Therefore, we can consider that ANN is the most new popular method in
AE signal analysis.

GA applications with AE signal analysis in machine condition monitoring and fault diagnosis
still need more support and atention because of the lack of existing evidence. The experi-
mental results prove that the use of fuzzy logic method is eicient and feasible. The eforts to
ind a novel idea must be encouraged to give more contributions in robust machine condition
monitoring and fault diagnosis.

Finally, the ability to continually change and obtain a novel idea for machine condition moni-
toring and diagnosis using AE signal analysis and AI will be in future works.

Acknowledgements

The author would like to thank Northern Technical University in Iraq through Professor Dr
Mowafaq Y. Hamdoon, the Chancellor of the university for supporting this work.

Conlict of interest

The author declares that there is no conlict of interest regarding the publication of this
chapter.
 Artificial Intelligence Application in Machine Condition Monitoring and Fault Diagnosis 287
http://dx.doi.org/10.5772/intechopen.74932

Author details

Yasir Hassan Ali

Address all correspondence to: [email protected]

Technical College Mosul, Northern Technical University, Mosul, Iraq

References

[1] Ali YH, Rahman RA, Hamzah RIR. Acoustic emission signal analysis and artiicial intel-
ligence techniques in machine condition monitoring and fault diagnosis: A review.
Journal Teknologi. 2014;69(2)

[2] Ali YH et al. Acoustic emission technique in condition monitoring and fault diagnosis of
gears and bearings. International Journal of Academic Research. 2014;6(5)
[3] Mba D, Rao RB. Development of Acoustic Emission Technology for Condition Monitoring
and Diagnosis of Rotating Machines; Bearings, Pumps, Gearboxes, Engines and Rotating
Structures. 2006
[4] Filippeti F, Franceschini G, Tassoni C. A survey of AI techniques approach for induc-
tion machine on-line diagnosis. Proceedings of Power Electronics and Motion Control
PEMC. 1996;2:314-318
[5] Siddique A, Yadava G, Singh B. 2003. Applications of artiicial intelligence techniques for
induction machine stator fault diagnostics: Review. Diagnostics for Electric Machines,
Power Electronics and Drives. 2003. SDEMPED 2003. 4th IEEE International Symposium
on: 29-34
[6] Kok JN, Boers EJ, Kosters WA, van der Puten P, Poel M. Artiicial intelligence: Deinition,
trends, techniques, and cases. Artiicial Intelligence. 2009
[7] Pham D, Pham P. Artiicial intelligence in engineering. International Journal of Machine
Tools and Manufacture. 1999;39(6):937-949
[8] Saxena A, Saad A. Genetic algorithms for artiicial neural net-based condition monitoring
system design for rotating mechanical systems. Applied Soft Computing Technologies:
The Challenge of Complexity. Springer. 2006. pp. 135-149
[9] Aguiar PR, Martins CH, Marchi M, Bianchi EC. Digital Signal Processing for Acoustic
Emission; 2012
[10] Kouroussis D, Anastassopoulos A, Lenain J, Proust A. Advances in Classiication of Acoustic
Emission Sources. Reims: Les Journées COFREND; 2001
[11] Al-Balushi K, Samanta B. Gear fault diagnostics using wavelets and artiicial neural
network. COMADEM 2000. 13th International Congress on Condition Monitoring and
Diagnostic Engineering Management. 1001-1010. 2000
288 Artificial Intelligence - Emerging Trends and Applications

[12] Mahamad AKB. Diagnosis, Classiication and Prognosis of Rotating Machine using
Artiicial Intelligence. Kumamoto University; 2010
[13] Abu-Mahfouz I. Condition monitoring of a gear box using vibration and acoustic emis-
sion based artiicial neural network. SAE Transactions. 2001;110(6):1771-1781
[14] Menon S, Schoess JN, Hamza R, Busch D. Wavelet-based acoustic emission detection
method with adaptive thresholding. SPIE's 7th Annual International Symposium on
Smart Structures and Materials. 2000. pp. 71-77
[15] Blahacek M, Chlada M, Prevorovský Z. Acoustic emission source location based on sig-
nal features. Advanced Materials Research. 2006;13:77-82
[16] Fog TL, Brown E, Hansen H, Madsen L, Sørensen P, Hansen E, Steel J, Reuben R, Pedersen
P. Exhaust Valve leakage Detection in Large Marine Diesel Engines. COMADEM´ 98,
11th Int. Conf. on Condition Monitoring and Diagnostic Engineering Management.
pp. 269-279
[17] Kouroussis D, Anastassopoulos A, Vionis P, Kolovos V. Unsupervised Patern recog-
nition of acoustic emission from full scale testing of a wind turbine blade. Journal of
Acoustic Emission (USA). 2000;18:217
[18] Wang J-Z, Wang L-S, Li G-f, Zhou G-H. Prediction of surface roughness in cylindrical
traverse grinding based on ALS algorithm. Machine Learning and Cybernetics, 2005.
Proceedings of 2005 International Conference on, 1. 2005: 549-554
[19] Wang Z, Willet P, DeAguiar PR, Webster J. Neural network detection of grinding
burn from acoustic emission. International Journal of Machine Tools and Manufacture.
2001;41(2):283-309
[20] Doto FR, Aguiar PR d, Bianchi EC, Serni PJ, Thomazella R. Automatic system for ther-
mal damage detection in manufacturing process with internet monitoring. Journal of the
Brazilian Society of Mechanical Sciences and Engineering. 2006;28(2):153-160
[21] Kwak J-S, Ha M-K. Neural network approach for diagnosis of grinding operation by
acoustic emission and power signals. Journal of Materials Processing Technology. 2004;
147(1):65-71
[22] Aguiar P, França T, Bianchi E. Roughness and roundness prediction in grinding.
Proceedings of the 5th CIRP International Seminar on Intelligent Computation in
Manufacturing Engineering (CIRP ICME 06). 2006. pp. 25-28
[23] Aguiar PR, Cruz CE, Paula WC, Bianchi EC. Predicting Surface Roughness in Grinding
using Neural Networks
[24] Goebel K, Wright PK. Monitoring and Diagnosing Manufacturing Processes Using a
Hybrid Architecture with Neural Networks and Fuzzy Logic. EUFIT, Proceedings. 1993. 2
[25] Walker JL, Russell SS, Workman GL, Hill EV. Neural network/acoustic emission burst
pressure prediction for impact damaged composite pressure vessels. Materials Evaluation.
1997;55(8):903-907
 Artificial Intelligence Application in Machine Condition Monitoring and Fault Diagnosis 289
http://dx.doi.org/10.5772/intechopen.74932

[26] Shen G, Duan Q, Zhou Y, Li B, Liu Q, Li C, Jiang S. Investigation of artiicial neural network
patern recognition of acoustic emission signals for pressure vessels. NDT. 2001;23:144-146

[27] Macías EJ, Roca AS, Fals HC, Fernández JB, Muro JC. Neural networks and acoustic
emission for modelling and characterization of the friction stir welding process. Revista
Iberoamericana de Automática e Informática Industrial RIAI. 2013;10(4):434-440

[28] Tian Y, Lewin P, Davies A, Suton S, Swingler S. Application of acoustic emission techniques
and artiicial neural networks to partial discharge classiication. Electrical Insulation,
2002. Conference Record of the 2002 IEEE International Symposium on: 119-123. 2002

[29] Szyszko S, Payne P. Artiicial neural networks for feature extraction from acoustic emis-
sion signals. Measurements, Modelling and Imaging for Non-Destructive Testing, IEE
Colloquium on. 1991:6/1-6/6
[30] Ali YH, Abd Rahman R, Hamzah RIR. Artiicial neural network model for monitoring
oil ilm regime in spur gear based on acoustic emission data. Shock and Vibration.
2015;2015

[31] Ali YH, Rahman RA, Hamzah RIR. Regression modeling for spur gear condition moni-
toring through oil ilm thickness based on acoustic emission signal. Modern Applied
Science. 2015;9(8):21
[32] Ahmed FY, Yusob B, Hamed HNA. Computing with spiking neuron networks a review.
International Journal of Advances in Soft Computing & Its Applications. 2014:6(1)
[33] Kasabov N. To spike or not to spike: A probabilistic spiking neuron model. Neural
Networks. 2010;23(1):16-19
[34] Kasabov N. Integrative connectionist learning systems inspired by nature: current mod-
els, future trends and challenges. Natural Computing. 2009;8(2):199-218
[35] Kojima, H. and S. Katsumata. An analysis of synaptic transmission and its plastic-
ity by glutamate receptor channel kinetics models and 2-photon laser photolysis. In:
International Conference on Neural Information Processing. 2008. Springer

[36] Ikegaya Y et al. Statistical signiicance of precisely repeated intracellular synaptic pat-
terns. PloS One. 2008;3(12):e3983

[37] Ahmed FY, Shamsuddin SM, Hashim SZM. Improved SpikeProp for using particle
swarm optimization. Mathematical Problems in Engineering. 2013;2013

[38] Izhikevich E.. Dynamical Systems in Neuroscience: The Geometry of Excitability and
Bursting (Computational Neuroscience): The MIT Press. 2006

[39] Izhikevich EM, Edelman GM. Large-scale model of mammalian thalamocortical sys-
tems. Proceedings of the National Academy of Sciences. 2008;105(9):3593-3598

[40] Kasabov, N. Evolving spiking neural networks and neurogenetic systems for spatio-and
spectro-temporal data modelling and patern recognition. In: IEEE World Congress on
Computational Intelligence. 2012. Springer
290 Artificial Intelligence - Emerging Trends and Applications

[41] Silva RG. Condition monitoring of the cuting process using a self-organizing spiking
neural network map. Journal of Intelligent Manufacturing. 2010;21(6):823-829

[42] Silva RG, Wilcox S, Araújo AA. Multi-sensor condition monitoring using spiking neuron
networks. In: IADIS international conference applied computing 2007. 2007

[43] Ming ZX. Application of Acoustic Emission Technique in Fault Diagnostics of Rolling
Bearing. Master's thesis. Tsinghua University, Beijing, Haidian; 2006

[44] Sibil A, Godin N, R’Mili M, Maillet E, Fantozzi G. Optimization of acoustic emission

data clustering by a genetic algorithm method. Journal of Nondestructive Evaluation.
2012;31(2):169-180
[45] Zadeh LA. Fuzzy sets. Information and Control. 1965;8(3):338-353

[46] Zadeh LA. Outline of a new approach to the analysis of complex systems and decision
processes. IEEE Transactions on Systems, Man and Cybernetics. 1973;(1):28-44

[47] Zadeh LA. Fuzzy algorithms. Information and Control. 1968;12(2):94-102

[48] Hellmann M. Fuzzy Logic Introduction” a Laboratoire Antennes Radar Telecom. FRE
CNRS. 2272

[49] Cusido J, Delgado M, Navarro L, Sala V, Romeral L. EMA fault detection using fuzzy
inference tools. AUTOTESTCON, 2010 IEEE. 2010: 1-6

[50] Omkar S, Suresh S, Raghavendra T, Mani V. Acoustic emission signal classiication using
fuzzy C-means clustering. Neural Information Processing, 2002. ICONIP'02. Proceedings
of the 9th International Conference on, 4. 2002:1827-1831

[51] De Aguiar PR, Bianchi EC, Canarim RC. Monitoring of Grinding Burn by Acoustic Emission

[52] Ren Q, Baron L, Balazinski M. Fuzzy identiication of cuting acoustic emission with
extended subtractive cluster analysis. Nonlinear Dynamics. 2012;67(4):2599-2608

[53] Ren Q, Baron L, Balazinski M. Application of type-2 fuzzy estimation on uncertainty in

machining: An approach on acoustic emission during turning process. Fuzzy Information
Processing Society, 2009. NAFIPS 2009. Annual Meeting of the North American. 2009: 1-6

[54] Ren Q, Baron L, Balazinski M. Type-2 fuzzy modeling for acoustic emission signal in
precision manufacturing. Modelling and Simulation in Engineering. 2011;2011:17

[55] Ren Q, Balazinski M, Baron L, Jemielniak K, Botez R, Achiche S. Type-2 fuzzy tool
condition monitoring system based on acoustic emission in micromilling. Information
Sciences. 2014;255:121-134
[56] Ren Q, Baron L, Balazinski M, Jemielniak K. Acoustic emission signal feature analysis
using type-2 fuzzy logic system. Fuzzy Information Processing Society (NAFIPS), 2010
Annual Meeting of the North American. 2010: 1-6

[57] Blahacek M, Prevorovsky Z, Krofta J, Chlada M. Neural network localization of noisy

AE events in dispersive media. Journal of Acoustic Emission(USA). 2000;18:279
 Artificial Intelligence Application in Machine Condition Monitoring and Fault Diagnosis 291
http://dx.doi.org/10.5772/intechopen.74932

[58] Khalifa S, Komarizadeh MH. An intelligent approach based on adaptive neuro-fuzzy

inference systems (ANFIS) for walnut sorting. Australian Journal of Crop Science.
2012;6(2)
[59] Vapnik V. Statistical Learning Theory New York. NY: Wiley; 1998

[60] Burges CJ. A tutorial on support vector machines for patern recognition. Data Mining
and Knowledge Discovery. 1998;2(2):121-167

[61] Saravanan N, Kumar Siddabatuni V, Ramachandran K. A comparative study on classii-

cation of features by SVM and PSVM extracted using Morlet wavelet for fault diagnosis
of spur bevel gear box. Expert Systems with Applications. 2008;35(3):1351-1366
[62] Widodo A, Yang B-S, Kim EY, Tan AC, Mathew J. Fault diagnosis of low speed bear-
ing based on acoustic emission signal and multi-class relevance vector machine.
Nondestructive Testing and Evaluation. 2009;24(4):313-328

[63] Widodo A, Kim EY, Son J-D, Yang B-S, Tan AC, Gu D-S, Choi B-K, Mathew J. Fault
diagnosis of low speed bearing based on relevance vector machine and support vector
machine. Expert Systems with Applications. 2009;36(3):7252-7261
[64] Yu Y, Zhou L. Acoustic emission signal classiication based on support vector machine.
TELKOMNIKA Indonesian Journal of Electrical Engineering. 2012;10(5):1027-1032
[65] Chu-Shu K. A machine learning approach for locating acoustic emission. EURASIP
Journal on Advances in Signal Processing. 2010;2010
[66] Yang Z, Yu Z. Grinding wheel wear monitoring based on wavelet analysis and sup-
port vector machine. The International Journal of Advanced Manufacturing Technology.
2012;62(1-4):107-121
[67] Yu Y, Zhou L. Acoustic emission signal classiication based on support vector machine.
Computer Engineering and Technology (ICCET), 2010 2nd International Conference,
16-18 April Chengdu. 2010;6:300-304

[68] Elforjani M, Shanbr S. Prognosis of bearing acoustic emission signals using supervised
machine learning. IEEE Transactions on Industrial Electronics. 2017

[69] Ali YH, Ali SM, Rahman RA, Hamzah RIR. Acoustic Emission and Artiicial Intelligent
Methods in Condition Monitoring of Rotating Machine–A Review. National Conference
for Postgraduate Research 2016, Universiti Malaysia Pahang. Malaysia. 2016