Available online at www.sciencedirect.

com

Expert Systems
with Applications
Expert Systems with Applications 36 (2009) 31193135
www.elsevier.com/locate/eswa

Vibration-based fault diagnosis of spur bevel gear box

using fuzzy technique
N. Saravanan *, S. Cholairajan, K.I. Ramachandran
Department of Mechanical Engineering, Amrita Vishwa Vidyapeetham, Coimbatore, Tamil Nadu 641105, India

Abstract
To determine the condition of an inaccessible gear in an operating machine the vibration signal of the machine can be continuously
monitored by placing a sensor close to the source of the vibrations. These signals can be further processed to extract the features and
identify the status of the machine. The vibration signal acquired from the operating machine has been used to eectively diagnose
the condition of inaccessible moving components inside the machine. Suitable sensors are kept at various locations to pick up the signals
produced by machinery and these signals are very meaningful in condition diagnosis surveillance. To determine the important characteristics and to unravel the signicance of these signals, further analysis or processing is required.
This paper presents the use of decision tree for selecting best statistical features that will discriminate the fault conditions of the gear
box from the signals extracted. These features are extracted from vibration signals. A rule set is formed from the extracted features and
fed to a fuzzy classier. The rule set necessary for building the fuzzy classier is obtained largely by intuition and domain knowledge.
This paper also presents the usage of decision tree to generate the rules automatically from the feature set. The vibration signal from a
piezo-electric transducer is captured for the following conditions good bevel gear, bevel gear with tooth breakage (GTB), bevel gear
with crack at root of the tooth (GTC), and bevel gear with face wear of the teeth (TFW) for various loading and lubrication conditions.
The statistical features were extracted and good features that discriminate the dierent fault conditions of the gearbox were selected using
decision tree. The rule set for fuzzy classier is obtained by once using the decision tree again. A fuzzy classier is built and tested with
representative data. The results are found to be encouraging.
2008 Elsevier Ltd. All rights reserved.
Keywords: Feature selection; Statistical features; Decision tree; Gear box; Fuzzy; Fault detection

1. Introduction
A faulty gear system could result in serious damage if
defects occur to one of the gears during operation condition. Early detection of the defects, therefore, is crucial to
prevent the system from malfunction that could cause damage or entire system halt. Diagnosing a gear system by
examining the vibration signals is the most commonly used
method for detecting gear failures. The conventional methods for processing measured data contain the frequency

Corresponding author. Tel.: +91 4222656422; fax: +91 4222656274.

E-mail addresses: n_saravanan@ettimadai.amrita.edu, nsaro_2000@
yahoo.com (N. Saravanan).
0957-4174/$ - see front matter 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.eswa.2008.01.010

domain technique, time domain technique, and time

frequency domain technique. These methods have been
widely employed to detect gear failures. The use of vibration analysis for gear fault diagnosis and monitoring has
been widely investigated and its application in industry is
well established (Cameron & Stuckey, 1994; Gadd &
Mitchell, 1984; Leblanc, Dube, & Devereux, 1990). This
is particularly reected in the aviation industry where the
helicopter engine, drive trains and rotor systems are tted
with vibration sensors for component health monitoring.
The raw vibration signal in any mode from a single point
on a machine is not a good indicator of the health or condition of a machine. Vibration is a vectorial parameter with
three dimensions and requires to be measured at several
carefully selected points.

3120

N. Saravanan et al. / Expert Systems with Applications 36 (2009) 31193135

Vibration Signals

Feature Selection
Using J 48 Algorithm

Rule Generation

Test data set

Modeling Fuzzy
system
Fuzzy inference engine

Fuzzy output

Bevel
Gear Box

Machine Condition Diagnosis

Sensor
Fig. 2. Fault simulator setup.

Fig. 1. Flowchart for bevel gear box health diagnosis.

Vibration analysis can be carried out using Fourier

transform techniques like Fourier series expansion (FSE),
Fourier integral transform (FIT) and discrete Fourier
transform (DFT) (Collacott, xxxx). After the development
of large-scale integration (LSI) and the associated microprocessor technology, fast Fourier transform (FFT) analyzers became cost eective for general applications. The
raw signatures acquired through a vibration sensor needed
further processing and classication of the data for any
meaningful surveillance of the condition of the system
being monitored.
Articial neural network (ANN), support vector
machine (SVM) and Fuzzy classier are widely used as
classication tool and reported in literature (Burgess,
1998; Jack & Nandi, 2000a; Nandi, 2000; Samanta & AlBaulshi, 2003; Samanta, Al-Baulshi, & Al-Araimi, 2003;
Shi et al., 1988). Among them, ANN has limitations on
generalization of the results in models that can over t
the data (Samanta et al., 2003). SVM has high classication
accuracy and good generalization capabilities for crisp data
(Burgess, 1998; Jack & Nandi, 2000a; Shi et al., 1988). In
the problem at hand, the nature of the fault itself is fuzzy
in nature. Fuzzy classier models the physical problem
under study more closely. The ow chart of the fault diagnostic system is shown in Fig. 1.
1.1. Dierent phases of present work
The signals obtained are processed further for machine
condition diagnosis as explained in the ow chart Fig. 1.
2. Experimental studies
The fault simulator with sensor is shown in Fig. 2 and
the pinion and gear shown in Fig. 3. A variable speed
DC motor (0.5 hp) with speed up to 3000 rpm is the basic

Electromagnetic
spring loaded disc
brake

Pinion
Wheel

Gear
Wheel

Fig. 3. Inner view of the bevel gear box.

drive. A short shaft of 30 mm diameter is attached to the

shaft of the motor through a exible coupling; this is to
minimize eects of misalignment and transmission of vibration from the motor. The shaft is supported at its ends
through two roller bearings. From this shaft the drive is
transmitted to the bevel gear box by means of a belt drive.
The gear box is of dimension 150 mm  170 mm  120 mm
and the full lubrication level is 110 mm and half lubrication
level is 60 mm.
SAE 40 oil was used as a lubricant. An electromagnetic
spring-loaded disc brake was used to load the gear wheel. A
torque level of 8 N-m was applied at the full-load condition. The various defects are created in the pinion wheels
and the mating gear wheel is not disturbed. With the sensor
mounted on top of the gear box vibrations signals are
obtained for various conditions. The selected area on the
top of the gearbox for mounting the sensor is made at
and smooth to ensure eective coupling between the sensor
and the gearbox. The sensor used is a piezoelectric accelerometer (Dytran model) which is mounted on the at

N. Saravanan et al. / Expert Systems with Applications 36 (2009) 31193135

surface using direct adhesive mounting technique. The

accelerometer is connected to the signal-conditioning unit
(DACTRAN FFT analyzer), where the signal goes through
the charge amplier and an analogue-to-digital converter
(ADC). The vibration signal in digital form is fed to the
computer through a USB port. The software RT Pro-series
that accompanies the signal-conditioning unit is used for
Table 1
Details of faults under investigation
Gears

Fault description

Dimension (mm)

G1
G2
G3
G4

Good
Gear tooth breakage (GTB)
Gear with crack at root (GTC)
Gear with face wear

8
0.8  0.5
0.5

Table 2
Gear wheel and pinion details
Parameters

Gear wheel

Pinion wheel

No. of teeth
Module
Normal pressure angle ()
Shaft angle ()
Top clearance (mm)
Addendum (mm)
Whole depth (mm)
Chordal tooth thickness (mm)
Chordal tooth height (mm)
Material

35
2.5
20
90
0.5
2.5
5.5
3.930.150
2.53
EN8

25
2.5
20
90
0.5
2.5
5.5
3.920.110
2.55
EN8

3121

recording the signals directly in the computers secondary

memory. The signal is then read from the memory and processed to extract dierent features.

2.1. Experimental procedure

In the present study, four pinion wheels whose details
are as mentioned in Table 1 were used. One was a new
wheel and was assumed to be free from defects. In the other
three pinion wheels, defects were created using electron discharge machine (EDM) in order to keep the size of the
defect under control. The details of the various defects
are depicted in Table 2 and its views are shown in Fig. 4.
The size of the defects is in-line with work reported in
literature (Gadd & Mitchell, 1984). The vibration signal
from the piezoelectric pickup mounted on the test bearing
was taken, after allowing initial running of the bearing for
sometime. The sampling frequency was 12,000 Hz and
sample length was 8192 for all speeds and all conditions.
The sample length was chosen arbitrarily, however, the following points were considered. Statistical measures are
more meaningful, when the number of samples is more.
On the other hand, as the number of samples increases
the computation time increases. To strike a balance, sample
length of around 10000 was chosen. In some feature extraction techniques, which will be used with the same data, the
number of samples is to be 2n. The nearest 2n10,000 is
8192 and hence, it was taken as sample length. Many trials
were taken at the set speed and vibration signal was stored

Fig. 4. (a)View of good pinion wheel. (b) View of pinion wheel with face wear (GFW). (c) View of pinion wheel with tooth breakage (GTB).

N. Saravanan et al. / Expert Systems with Applications 36 (2009) 31193135

-0.2
6000

-0.4
-0.4

8000

2000

4000

6000

-0.2

8000

-0.2
4000

6000

0
-0.2
2000

2000

4000

6000

Amplitude

-0.1

0
-0.1

8000

Sample No.
GTC-HalfLub-Unload

2000

4000

6000

-0.1
4000

6000

-0.1

8000

2000

4000

6000

-0.1
0

2000

4000

Sample No.

6000

8000

-0.1
2000

4000

6000

2000

8000

Sample No.

4000

6000

8000

2000

4000

6000

0
-0.2
-0.4

8000

2000

TFW-Dry-Unload
0

4000

6000

0
-0.2
0

-0.2
6000

-0.2
4000

6000

Sample

6000

8000

0
-0.2
0

2000

4000

6000

8000

Sample
TFW-FullLub-FullLoad

2000

4000

0.2

8000

0.2

2000

Sample
TFW-HalfLub-FullLoad

4000

8000

0.2

8000

0.2

2000

6000

TFW-Dry-FullLoad

-0.2
2000

4000

Sample No.

Sample
TFW-FullLub-Unload

Sample No.
GTB-FullLub-FullLoad

0.2

Amplitude

Amplitude

0
-0.2
-0.4

8000

-0.2

8000

0.1

8000

0.2

Sample No.
GTC-FullLub-FullLoad

0.1

6000

6000

0.2

Sample
TFW-HalfLub-Unload

Sample No.
GTC-FullLub-Unload

4000

Amplitude

Amplitude

2000

4000

Sample No.
GTB-HalfLub-FullLoad

0.2

8000

0.1

2000

Sample No.
GTC-HalfLub-FullLoad

0.1

-0.4

2000

Sample No.

GTC-Dry-fullLoad

Sample No.
GTB-FullLub-Unload

8000

0.1
Amplitude

Amplitude

6000

0
-0.2
-0.4

8000

-0.2

Sample No.

GTC-Dry-Unload

Amplitude

4000

6000

-0.4
0

8000

4000

0.2

8000

0.2

Sample No.

c 0.1

6000

Amplitude

Amplitude

Amplitude

2000

4000

2000

0.2

Sample No.
GTB-HalfLub-Unload

Sample No.
Good-Full-FullLoad

0.2

2000

8000

Sample No.
Good-FullLub-Unload

Amplitude

6000

0.2

-0.2
-0.4

4000

Sample No.
Good-HalfLub-FullLoad

0.2
Amplitude

Amplitude

Sample No.
Good-HalfLub-Unload

2000

Amplitude

4000

0
-0.2

Amplitude

2000

-0.2

Amplitude

GTB-Dry-FullLoad

Amplitude

GTB-Dry-Unload
0.2

Amplitude

0.2
Amplitude

Amplitude

0.2

-0.4

Good-Dry-FullLoad

Amplitude

Good-Dry-Unload

Amplitude

Amplitude

3122

8000

0.2
0
-0.2
0

2000

4000

6000

8000

Sample

Fig. 5. (a) Vibration signal for good pinion wheel under dierent lubrication and loading conditions. (b) Vibration signal for pinion wheel with teeth
breakage under dierent lubrication and loading conditions. (c) Vibration signal for pinion wheel with crack at root under dierent lubrication and
loading conditions. (d) Vibration signals for pinion wheel with teeth face wear under dierent lubrication and loading conditions.

in the data le. The raw vibration signals acquired for various experimental conditions form the gearbox using FFT
are shown in Fig. 5.

conditions and fed as an input to J 48 algorithm for selecting the best features which classify the dierent fault
conditions.

3. Feature extraction

4. Descriptive statistics

Statistical analysis of vibration signals yields dierent

primary and secondary Parameters. Research work
reported (James Li & Wu, 1989) use these in combinations
to elicit information regarding bearing faults. Such procedures use allied logic often based on physical Considerations. A fairly wide set of these parameters is selected as
a basis for our study, as detailed below.
a. Mean b. Standard error c. Median d. Standard deviation e. Sample variance f. Kurtosis g. Skewness h. Range
i. Minimum j. Maximum k. Sum.
All the above mentioned statistical features were
extracted for the vibration signals obtained for various

The statistical features are explained below.

4.1. Standard deviation
This is a measure of the eective energy or power content of the vibration signal and clearly indicates deterioration in the bearing condition. The following formula was
used for computation of standard deviation
s
P
P 2
n x2  x
:
Standard deviation
nn  1

N. Saravanan et al. / Expert Systems with Applications 36 (2009) 31193135

3123

4.2. Skewness

4.8. Maximum value

Skewness characterizes the degree of asymmetry of a

distribution around its mean. The below shown expression
was used to calculate the skewness, where n is the sample
size and s is the sample standard deviation
X xi  x3
n
:
Skewness
n  1n  2
s

It refers to the maximum signal point value in a given

signal.
4.9. Sum
It is the sum of all signal point values in a given signal.
5. Using J 48 algorithm in the present work

4.3. Kurtosis
Kurtosis indicates the atness or the spikiness of the signal. Its value is very low for good bevel gearbox and high
for faulty gearbox due to the spiky nature of the signal
(
Kurtosis

)
X xi  x4
nn 1
n  1n  2n  3
s
2

3n  1
:
n  2n  3

where s is the sample standard deviation.

4.4. Standard error
Standard error is a measure of the amount of error in
the prediction of y for an individual x in the regression,
where x and y are the sample means and n is the sample
size.
Standard error of the predicted
v
"
#
u
P
2
X
u 1



x

x
y

y


2
y  y 
:
yt
P
2
n  2
x  x
4.5. Sample variance
It is variance of the signal points and the following formula was used for computation of standard variance.
Sample variance

P 2
x2  x
:
nn  1

4.6. Range
It refers to the dierence in maximum and minimum signal point values for a given signal.

A standard tree induced with c5.0 (or possibly ID3 or

c4.5) consists of a number of branches, one root, a number
of nodes and a number of leaves. One branch is a chain of
nodes from root to a leaf; and each node involves one attribute. The occurrence of an attribute in a tree provides the
information about the importance of the associated attribute as explained by Peng, Flach, Brazdil, and Soares
(2002). A Decision Tree is a tree based knowledge representation methodology used to represent classication rules.
J48 algorithm (A WEKA implementation of c4.5 Algorithm) is a widely used one to construct Decision Trees
as explained by Sugumaran, Muralidharan, and Ramachandran (2006).
The Decision Tree algorithm has been applied to the
problem under discussion. Input to the algorithm is set of
statistical features of vibration signatures. It is clear that
the top node is the best node for classication. The other
features in the nodes of Decision Tree appear in descending
order of importance. It is to be stressed here that only features that contribute to the classication appear in the
Decision Tree and others do not. Features, which have less
discriminating capability, can be consciously discarded by
deciding on the threshold. This concept is made use for
selecting good features. The algorithm identies the good
features for the purpose of classication from the given
training data set, and thus reduces the domain knowledge
required to select good features for pattern classication
problem. The decision trees shown in Figs. 7, 13, 18, 22,
27 and 31 are for various lubrication and loading conditions of dierent faults compared with good conditions of
the pinion gear wheel.
Based on the output of J 48 algorithm, the decision tree
various statistical parameters are selected for the various
conditions of the gearbox. The values appearing between
various nodes in the decision tree are used for generating
the fuzzy rules to classify the various conditions of the
gearbox under study.
5.1. Application of decision tree for feature selection

4.7. Minimum value

It refers to the minimum signal point value in a given
signal. As the gear parts (crack, breakage, face wear) get
degraded, the vibration levels seem to go high. Therefore,
it can be used to detect faulty gears.

The algorithm has been applied to the problem under

discussion for feature selection. Input to the algorithm is
the set of statistical features described in Section 3
extracted from raw vibration signatures, the output is the
Decision Tree. It is clear there from that the top node is
the best node for classication. The other features appear

3124

N. Saravanan et al. / Expert Systems with Applications 36 (2009) 31193135

in the nodes in Decision Tree in descending order of importance. It is to be stressed here that only features that contribute to the classication appear in the Decision Tree
and others do not. The level of contribution is not same
and all statistical features are not equally important. The
level of contribution by individual feature is given by a statistical measure within the parenthesis in the Decision Tree.
The rst number in the parenthesis indicates the number of
data points that can be classied using that feature set. The
second number indicates the number of samples against
this action. If the rst number is very small compared to
the total number of samples, then the corresponding features can be considered as outliers and hence ignored. Features that have less discriminating capability can be
consciously discarded by deciding on the threshold. This
concept is made use of in selecting good features. The algorithm identies the good features for the purpose of classication from the given training data set and thus reduces
the domain knowledge required to select good features
for pattern classication problem.
6. Methodology adopted for fuzzy classication
See Fig. 6.
7. Fuzzy logic (classier)
Fuzzy Logic provides a precise approach for dealing
with uncertainty. Fuzzy inference is a method that interprets the values in the input vector and, based on some
set of rules, assigns values to the output vector. The
point of fuzzy logic is to map an input space to an output space, and the primary mechanism for doing this is a
list of if-then statements called rules. Rules are the
inputs for building a fuzzy inference engine. The methodology adopted for fuzzy classication is shown in Fig. 6.
All rules are evaluated in parallel, and the order of the
rules is unimportant. The real world data do not have
sharply dened boundaries where information is often
incomplete or sometimes unreliable. In quest for precision, scientists have generally attempted to manipulate
the real world into articial mathematical models that
make no provision for gradation. Because Fuzzy Logic
provides the tools to classify information into broad,
coarse categorizations or groupings, it has innite possi-

bilities for application which have proven to be much

cheaper, simpler and more eective than other systems
in handling complex information (Cox, 1994).
For the problem at hand, the condition of the gearbox,
good or faulty is basically fuzzy in nature. All the faults do
not occur in the gearbox instantly. It comes gradually. In
that case, there is no threshold value (crisp data) based
on which the decision on the condition of the gearbox
can be taken. (Whether gearbox is now good or faulty).
The problems of this kind can be modeled using fuzzy logic
more closely (Huang, 1997; Rao, 1996; Zeng & Wang,
1991).
8. Membership function
A membership function (MF) is a curve that denes how
each point in the input space is mapped to a membership
value (or degree of membership) between 0 and 1. Observing the values of the feature, based on which the branches
of the Decision Tree are created for dierent conditions of
the gearbox, the membership functions for the corresponding features are dened. There are four possible outcomes
from a fuzzy classier, namely: good bevel gear, bevel gear
with tooth breakage (GTB), bevel gear with crack at root
of the tooth (GTC), and bevel gear with face wear of the
teeth (TFW) for various loading and lubrication conditions. Hence, four membership functions are dened with
equal range for the output.
9. Rule generation from decision tree
Articial neural network and support vector machine
are used to generate rule for classication problems
(Andrews et al., 1995; Haydemar, Cecilio, & Andreu,
xxxx). In this study, Decision Tree is used for that purpose. Decision Tree shows the relation between features
and the condition of the gearbox. Tracing a branch from
the root node leads to a condition of the gearbox (Refer
Figs. 7, 13, 18, 22, 27 and 31) and decoding the information available in a branch in the form of if-then statement gives the rules for classication using fuzzy for

Dry/Half/Full

No Load

Full Load

GOOD/GTC/
GTB/TFW

Fig. 6. Methodology of classication using Fuzzy.

Fig. 7. Decision tree from J 48 Algorithm for dry-lub no-load condition.

N. Saravanan et al. / Expert Systems with Applications 36 (2009) 31193135

various conditions of the gearbox. Hence the usefulness of

the decision tree in forming the rules for fuzzy classication is established.
10. Generation of rules for various gearbox conditions and
discussions
The preceding section describes how the classication
has been carried out using fuzzy technique.

3125

10.1. Dry-lubrication and no-load condition

From Fig. 7 we can see that standard error, kurtosis and
variance play a decisive role in classifying the various gearbox faults under dry lubrication and no-load condition.
This output of the decision tree is used to design the membership function for fuzzy classier as shown in Fig. 810.
A membership function (MF) is a curve that denes how
each point in the input space is mapped to a membership

Fig. 8. Membership function for standard error.

Fig. 9. Membership function for kurtosis.

Fig. 10. Membership function for variance.

3126

N. Saravanan et al. / Expert Systems with Applications 36 (2009) 31193135

Fig. 11. Membership functions for condition. (output).

value (or degree of membership) between 0 and 1. In the

present study, trapezoidal membership function is used.
The selection of this membership function is to some extent
arbitrary. However, the following points were considered
while selecting membership function. The Decision Tree
for the selected three features is shown in Fig. 7. Observing
the values of the feature, based on which the branches of
the Decision Tree is created, the membership functions
for all three features are dened for standard error, kurtosis and variance, respectively.
10.1.1. Rules designed for the dry-lubrication and no-load
condition
Figs. 1416.
1. If (stderr is not stderr) then (Output1 is GTC).
2. If (stderr is stderr) and (kurtosis is Kur) then (Output1 is
GOOD).
3. If (stderr is stderr) and (kurtosis is not Kur) and (variance is Var) then (Output1 is GTB).
4. If (stderr is stderr) and (kurtosis is not Kur) and (variance is not Var) then (Output1 is TFW).
The membership value of the condition being GTC is
when the standard error value is less than or equal to
0.000175 (from Fig. 7) which is the threshold value. Hence,
up to this threshold value the membership function generates the value 0 and afterwards it increases linearly
(assumption). The trapezoidal membership function suits
this phenomenon and hence it was selected to map each
point in the input space to a membership value. To review,
the threshold values are given by decision tree and the slope
is dened by the user through heuristics. The threshold
value (0.000175) is dened based on the representative
training dataset. If standard error value is less than or
equal to 0.000175, a membership function which is dened
on a 01 scale gives a value of 0 which means that it is not a
standard error. If threshold value is greater than 0.000175,
the membership function generates a value of 1. Similarly
membership functions for other features are designed
accordingly and shown in Figs. 9 and 10.

There are four possible outcomes from a fuzzy classier,

namely: Good, GTC, GTB and TFW Hence, four membership functions are dened with equal range and shown in
Fig. 11.
10.1.2. Fuzzy inference engine
After dening membership functions and generating the
if-then rules, the next step is to build the fuzzy inference
engine. The fuzzy toolbox available in MATLAB 7 was
used for building fuzzy inference engine. Each rule was
taken at a time and using membership functions and fuzzy
operators the rules were entered. The rules were obtained
from a training data set (150 trials in each condition).
For testing the built model a portion of the data (100 trials
in each condition) called testing data was kept aside. Using
the testing data, the fuzzy inference engine was evaluated
and its performance was presented as confusion matrix in
Table 3. The diagonal elements in the confusion matrix
table (3) show the number of correctly classied instances.
In the rst row, the rst element shows the number of data
points belonging to good class and classied by fuzzy
logic as good. The second element shows the number of
data points belonging to GTC class and classied by fuzzy
logic as GTC. The third element shows the number of
data points belonging to GTB class and classied by fuzzy
logic as GTB. The fourth element shows the number of
data points belonging to TFW class and classied by
fuzzy logic as TFW. Table 3 illustrates the powerfulness
of the fuzzy rules designed with the aid of the decision trees
by the authors.
Fig. 12 illustrates the application of the rules designed.
Here each row corresponds to each rule as discussed in

Table 3
Condition

GOOD

GTC

GTB

TFW

GOOD
GTC
GTB
TFW

100
0
0
0

0
100
0
0

0
0
100
0

0
0
0
100

N. Saravanan et al. / Expert Systems with Applications 36 (2009) 31193135

3127

Fig. 12. Rule viewer for one of the test data.

section 6.4.2. The rst three blocks in rows represents the

membership function of standard error, kurtosis, variance,
respectively. The fourth block corresponds to the membership functions for output as shown in Fig. 11. With the
help of sample inputs for standard error, kurtosis and variance the rules are tested as follows, for a sample input of
standard error as 0.0005, kurtosis as 10 and variance as
0.005 which satises the second rule completely and the
corresponding output condition is GOOD, which is shown
in the output block of the second row in the rule viewer
shown in Fig. 12.

10.2.1. Rules designed for the dry-lubrication and full-load

condition
See Figs. 1416.

10.1.3. Confusion matrix

In lieu with the above discussions the fuzzy rules, membership functions, confusion matrix and rule viewer are
shown in Sections 10.210.5 and 10.6.

Here there are three membership functions to represent

three threshold values of variance in the decision tree.
Fig. 17 is the rule viewer for the following test data.If
stderr = 0.0005, kurtosis = 10, Variance = 0.005, then the
output is 6.25, i.e., the condition is GTB.

1. If (stderr is stderr) then (output1 is GTC).

2. If (stderr is not stderr) and (variance is var3) then (output1 is GTB).
3. If (kurtosis is kur) and (variance is var2) then (output1 is
GOOD).
4. If (kurtosis is not kur) and (variance is var1) then (output1 is TFW).

10.2. Dry-lubrication and full-load condition

See Fig. 13.

10.2.2. Confusion matrix

Table 4.

Fig. 13. Decision tree from J 48 Algorithm for dry-lub full-load condition.

3128

N. Saravanan et al. / Expert Systems with Applications 36 (2009) 31193135

Fig. 14. Membership function for stderr.

Fig. 15. Membership function for kurtosis.

Fig. 16. Membership functions for variance.

10.3. Half-lubrication and no-load condition

4. If (stderr is stderr2) and (kurtosis is not kur) then (output1 is TFW).

See Fig. 18.

10.3.2. Confusion matrix
Table 5.
10.3.1. Rules designed for half -lubrication and no-load
condition
See Figs. 1921.

10.4. Half-lubrication and full-load condition

See Fig. 22.

1. If (stderr is stderr1) then (output1 is GTC).

2. If (stderr is not stderr2) then (output1 is GOOD).
3. If (stderr is stderr2) and (kurtosis is kur) then (output1 is
GTB).

10.4.1. Rules designed for half-lubrication and full-load

condition
Figs. 2326.

N. Saravanan et al. / Expert Systems with Applications 36 (2009) 31193135

3129

Fig. 17. Rule viewer for one of the test data.

Table 4
Condition

GOOD

GTC

GTB

TFW

GOOD
GTC
GTB
TFW

100
0
0
0

0
100
0
0

0
0
100
0

0
0
0
100

1. If (stderr is stderr1) then (output1 is GTC).

2. If (stderr is not stderr2) then (output1 is GOOD).
3. If (stderr is stderr2) and (kurtosis is not kur) then (output1 is TFW).
4. If (stderr is stderr2) and (kurtosis is kur) and (minimum
is min) then (output1 is GTB).

Fig. 18. Decision tree from J 48 Algorithm for half-lub no-load condition.

Fig. 19. Membership functions for stderr.

3130

N. Saravanan et al. / Expert Systems with Applications 36 (2009) 31193135

Fig. 20. Membership function for kurtosis.

Fig. 21. Rule viewer for one of the test data If stderr = 0.0005, kurtosis = 10. The output is 6.25, hence the condition is GTB.
Table 5
Condition

GOOD

GTC

GTB

TFW

GOOD
GTC
GTB
TFW

100
0
0
0

0
100
0
0

0
0
100
0

0
0
0
100

3. If (stderr is stderr2) and (variance is var) then (output1

is GTB).
4. If (stderr is stderr2) and (variance is not var) then (output1 is TFW).

10.5.2. Confusion matrix

Table 7.
5. If (stderr is stderr2) and (kurtosis is kur) and (minimum
is not min) then (output1 is TFW).

10.6. Full-lubrication and full-load condition

See Fig. 31.

10.4.2. Confusion matrix

Table 6.
10.5. Full-lubrication and no-load condition

10.6.1. Rules designed for full-lubrication and full-load

condition
Figs. 32 and 33.

See Fig. 27.

10.5.1. Rules designed for full-lubrication and no-load
condition
Figs. 2830.

1.
2.
3.
4.

If
If
If
If

(stderr
(stderr
(stderr
(stderr

is
is
is
is

stderr1) then (output1 is GTC).

not stderr3) then (output1 is GOOD).
stderr2) then (output1 is GTB).
stderr3) then (output1 is TFW).

1. If (stderr is stderr1) then (output1 is GTC).

2. If (stderr is not stderr2) then (output1 is GOOD).

10.6.2. Confusion matrix

Table 8.

N. Saravanan et al. / Expert Systems with Applications 36 (2009) 31193135

Fig. 22. Decision tree from J 48 Algorithm for half-lub full-load condition.

Fig. 23. Membership functions for stderr.

Fig. 24. Membership functions for kurtosis.

3131

3132

N. Saravanan et al. / Expert Systems with Applications 36 (2009) 31193135

Fig. 25. Membership functions for minimum.

Fig. 26. Rule viewer for one of the test data. If Stderr = 0.0005, kurtosis = 10, minimum = 0, output1 = 6.25, i.e., the condition is GTB.

Table 6

11. Conclusion

Condition

GOOD

GTC

GTB

TFW

GOOD
GTC
GTB
TFW

100
0
0
0

0
100
0
0

0
0
100
0

0
0
0
100

Fault diagnosis of Gear box is one of the core research

areas in the eld of condition monitoring of rotating
machines. The work conducted, proposing the method of

Fig. 27. Decision tree from J 48 Algorithm for full-lub no-load condition.

N. Saravanan et al. / Expert Systems with Applications 36 (2009) 31193135

3133

Fig. 28. Membership function for stderr.

Fig. 29. Membership function for variance.

Fig. 30. Rule viewer for one of the test data.If stderr = 0.0005, variance = 0.005, then output1 = 6.25.

Table 7
Condition

GOOD

GTC

GTB

TFW

GOOD
GTC
GTB
TFW

96
0
10
0

0
100
0
0

0
0
90
0

4
0
0
100

the gearbox fault identication based on fuzzy logic technique, shows the great potentiality and the strong ability
to classify and identify machinery faults. This work has
investigated the use of basic fuzzy logic principle as a fault
diagnostic technique for spur bevel gearbox. The work conducted has demonstrated the potential of fuzzy logic to

3134

N. Saravanan et al. / Expert Systems with Applications 36 (2009) 31193135

Fig. 31. Decision tree from J 48 Algorithm for full-lub full-load condition.

Fig. 32. Membership function for stderr.

Fig. 33. Rule viewer for one of the test data. If stderr = 0.0002 then output1 = 1.25, ie, the condition is Good.

N. Saravanan et al. / Expert Systems with Applications 36 (2009) 31193135

Table 8
Condition

GOOD

GTC

GTB

TFW

GOOD
GTC
GTB
TFW

100
0
0
0

0
100
0
0

0
0
100
0

0
0
0
100

classify the likely fault conditions which are represent in

rotating machinery. From the study carried out and presented in this paper, the diagnosis technique based on the
fuzzy logic principle is found to be practical for the condition recognition of the gearbox. Also this work brings out
the potential of decision trees to generate the rules automatically from the feature set which proves to be a great
asset in generating fuzzy rules. This work has outlined
the procedure of fuzzy diagnosis technique by using the
characteristic variables which represent a particular running condition of the gearbox to determine the fuzzy membership function.
References
Andrews, R., Diederich, J., & Tickle, A. (1995). A survey and critique of
techniques for extracting rules from trained articial neural networks.
Knowledge-Based Systems, 8(6), 373389.
Burgess, C. J. C. (1998). A tutorial on support vector machines for pattern
recognition. Data Mining and Knowledge Discovery, 2, 955974.
Cameron, B. G., & Stuckey, M. J. (1994). A review of transmission
vibration monitoring at Westland Helicopter Ltd. In Proceedings of the
20th European rotorcraft forum (pp. 16/1116/16) paper 116.
Collacott, R. A. Mechanical fault diagnosis and condition monitoring.
Chapman & Hall.
Cox, E. (1994). The fuzzy systems handbook a practitioners guide to
building, using, and maintaining fuzzy systems. New York: Academic
Press.
Gadd, P., & Mitchell, P. J. (1984). Condition monitoring of helicopter
gearboxes using automatic vibration analysis techniques, AGARD CP
369. Gears and power transmission system for helicopter turboprops, 29/
129/10.

3135

Haydemar, N., Cecilio, A., & Andreu C. Rule extraction from support
vector machines. In ESANN2002 proceedings European symposium
on articial neural networks (pp. 107112).
Huang, Y. C., Yang, H. T., & Huang, C. L. (1997). Developing a new
transformer fault diagnosis system through evolutionary fuzzy logic.
IEEE Transactions on Power Delivery 12.
Jack, L. B., & Nandi, A. K. (2000). Comparison of neural networks and
support vector machines in condition monitoring application. In
Proceedings of COMADEM 2000 (pp. 721730). Houston, TX, USA.
James Li, C., & Wu, S. M. (1989). Online detection of localized defects in
bearing by pattern recognition analysis. ASME Journal of Engineering
Industry, 111, 331336.
Leblanc, J. F. A., Dube, J. R. F., & Devereux, B. (1990). Helicopter
gearbox vibration analysis in the Canadian forces applications and
lessons. In Proceedings of the rst international conference, gearbox
noise and vibration (pp. 173177). IMechE Cambridge, UK, C404/023.
Nandi, A. K. (2000). Advanced digital vibration signal processing for
condition monitoring. In Proceedings of COMADEM2000 (pp. 129
143). Houston, TX, USA.
Peng, Y. H., Flach, P. A., Brazdil, P., & Soares, C. (2002). Decision treebased data characterization for meta-learning. In ECML/PKDD-2002
workshop IDDM-2002. Helsinki, Finland.
Rao, B. K. N. (1996). The handbook of condition monitoring. London:
Elsevier.
Samanta, B., & Al-Baulshi, K. R. (2003). Articial neural network
based fault diagnostics of rolling element bearings using time
domain features. Mechanical Systems and Signal Processing, 17(2),
317328.
Samanta, B., Al-Baulshi, K. R., & Al-Araimi, S. A. (2003). Articial
neural networks and support vector machines with genetic algorithm
for bearing fault detection. Engineering Applications of Articial
Intelligence, 16, 657665.
Shi, X. Z., Xu, Z. Q. & Xu. (1988). A study on the automatic recognition
of vibration signal for ball bearing faults The FFTAR feature
extraction and classication methods. In Proceedings of IEEE international workshop on applied time series analysis (pp. 318321). World
Scientic.
Sugumaran, V., Muralidharan, V., & Ramachandran, K. I. (2006).
Feature selection using decision tree and classication through
proximal support vector machine for fault diagnostics of roller
bearing. Mechanical Systems and Signal Processing, 21, 930942.
Zeng, L., & Wang, Z., (1991). Machine-fault classication: A fuzzy
approach. The International Journal of Advanced Manufacturing
Technology 6, 8394.