Energy and Buildings 55 (2012) 585591

Contents lists available at SciVerse ScienceDirect

Energy and Buildings

journal homepage: www.elsevier.com/locate/enbuild

A semi-automatic approach for thermographic inspection of electrical

installations within buildings
A.S. Nazmul Huda a , Soib Taib a, , Mohd Shawal Jadin b , Dahaman Ishak a
a
b

School of Electrical and Electronic Engineering, Universiti Sains Malaysia, 14300 Nibong Tebal, Pulau Pinang, Malaysia
Faculty of Electrical & Electronic Engineering, Universiti Malaysia Pahang, 26600 Pekan, Pahang, Malaysia

a r t i c l e

i n f o

Article history:
Received 20 June 2012
Received in revised form 4 September 2012
Accepted 17 September 2012
Keywords:
Electrical installation
Infrared image
Multilayered perceptron
Statistical features
Conditions classication
Building

a b s t r a c t
Maintaining the reliability of electrical installation has become part of the energy efciency practices in
building. The degradation of electrical installations can cause overheating, which can lead to subsequent
failure of the equipments that can potentially result in unplanned power outages, possible injury and
re hazard. In addition, the efciency of an electrical system becomes low prior to failure, thus energy
is spent generating heat and causing unnecessary energy loses. Therefore, early prevention is required
to avoid this situation by monitoring the reliability of the electrical installations through energy audit
practices. This article proposes a semi-automatic approach for evaluating the thermal condition of electrical installations within the building in Malaysia by analyzing its infrared image. Initially the interest
regions of the images are manually segmented. Then the statistical features of rst order histogram and
gray level co-occurrence matrix features as well as the differences of feature parameters between hot
and reference regions are extracted from segmented regions. Principle component analysis is applied for
the best features selection and at the nal stage, the condition of electrical equipments will be classied
using multilayered perceptron neural network. The performances of multilayered perceptron networks
have been compared and tested with various training algorithms. The classication accuracy of multilayered perceptron networks are also compared with discriminant analysis classier and it is found
that the multilayered perceptron network using LevenbergMarquardt algorithm gives the best testing
performance. The result shows that the maximum testing accuracy 78.5% was obtained.
2012 Elsevier B.V. All rights reserved.

1. Introduction
According to the latest report [1] of Fire and Rescue Department
of Malaysia about the causes of re in buildings, 2317 re related
incidents have occurred between January 2012 and June 2012, making the average number of incidents around 387 a month. In this
report, 1049 of res were caused by electrical problems which was
about 46% of the total causes of re in building and were mainly
involved the electrical wiring problems (809 cases) and failure of
electrical equipments (240 cases). Failure in electrical distribution
equipments can potentially produce an ignition and re. One of
the causes of ignition is excessive ohmic heating in electrical distribution [2]. This condition can occur especially for old buildings
with outdated electrical wiring that is deteriorating, inappropriately amended, or insufcient for the electrical loads. However,
new constructions also are not protected from this condition. If
the quality of the connection degrades, in effect, more energy will
dissipate from the devices as its electrical resistance increases [3].

Corresponding author. Tel.: +60 45996012.

E-mail address: soibtaib@eng.usm.my (S. Taib).
0378-7788/$ see front matter 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.enbuild.2012.09.014

Therefore, periodical monitoring and diagnosis of equipments condition are essential for early fault detection and maintaining the
energy efciency in buildings.
Infrared thermography (IRT) is the best and useful tool to
monitor and determine the heat related problems [48]. Various problems can also be detected within the monitoring such as
poor connections, short-circuits, overloads, load imbalances, and
improperly installed electrical components [9]. The system allows
on-line maintenance process without interruption of service, minimizes downtime, reduces outage and manpower cost, avoids
sudden failure of the equipments that could be catastrophic, injuring and losing of life [1012].
Applying automatic condition monitoring system can improve
the fault detection technique and the level of abnormalities in
electrical equipments can be evaluated even when the expert
or experienced personnel are not present [13,14]. For instance,
Almeida et al. proposed an intelligent faults diagnosis system based
on thermography for surge arrestors using two kinds of variables
as inputs of neuro-fuzzy network [15]. This system was designed to
classify faults into two classes which are normal or faulty and the
index of wrong classication result is found to be lower than 10%.
Shai and Hamzah used RGB colour scale data and temperature

586

A.S. Nazmul Huda et al. / Energy and Buildings 55 (2012) 585591

Table 1
Classication of conditions of electrical equipments.
Priority

Type of condition

T between similar components ( C)

Recommended action

1
2
3

Critical
Warning
Normal

T 15
5 < T < 15
T 5

Major discrepancy; repair immediately

Indicates probable deciency; repair as time permits
Minor overheating; warrants investigation

data as the input features of articial neural network to detect faults

in electrical equipments [16]. Rahmani et al. developed an intelligent system to detect faults of electrical equipments in ground
substations using support vector machine (SVM) as a classier and
22 image features of Zernike moments [17].
In this research work, an automatic classication system for
determining the conditions of electrical equipments was implemented. Initially the infrared images of electrical equipments were
manually segmented into two different region of interest (ROI) for
the same load and condition. One of the ROI is the suspected faulty
component while the other one is the normal component which is
usually assigned as the reference component. Most of the electrical
equipments and components are installed side by side; therefore it
is easy to segment all the components with the same size. Asymmetry analysis of these two ROIs is found to be the appropriate
way to analyse the conditions of the equipments. Two kinds of statistical based feature extraction methods namely histogram based
rst order statistical and second order grey level co-occurrence
matrix (GLCM) statistical texture analysis method are widely used
to describe the asymmetry between these two regions [18]. Seven
orders of the rst order statistical approach and four orders of
the GLCM texture analysis method were used. Total 22 features
are extracted for both reference and hot component as a separate
parameter namely reference and hot parameter respectively.
Suitable feature selection is an important task for ensuring better performance of classier especially in high building. Because,
some input features are highly uncorrelated or irrelevant those
can decrease the classier performance and this can cause incorrect analysis of energy editing process. For reducing the features,
various feature selection methods such as principle component
analysis (PCA), and discriminant analysis (DA) can be used. In
this article, PCA [19] is used to select the suitable features for
condition monitoring of equipments. It is found that PCA is the
simplest, easy to apply and provide true eigenvector based multivariate analysis tool. At the nal stage, multilayered perceptron
(MLP) neural networks are used for classifying the conditions
of electrical equipments which were divided into 3 classes i.e.
normal, warning and critical. The performance of MLP networks
were compared using twelve different training algorithms namely
LevenbergMarquardt (LM), BroydenFletcherGoldfarbShanno
quasi-Newton (BFG), resilient back propagation (RP), scaled conjugate gradient (SCG), conjugate gradient with Powell-Beale restarts
(CGB), Conjugate gradient with Fletcher-Reeves updates (CGF),
conjugate gradient with Polak-Ribiere updates (CGP), One step
secant (OSS), Bayesian regularisation (BR), Gradient descent (GD),
Gradient descent with momentum and adaptive learning rate
(GDX) and Gradient descent with momentum (GDM) algorithm.
Details of these algorithms can be found in [20]. The performances
of classication are evaluated based on overall percentage accuracy. Performance of MLP neural network is also compared with
discriminant analysis classier.
2. Infrared image acquisition and manual classication
approach
Infrared images of electrical equipments are captured from main
switch boards (MSB) from different locations of old ofce buildings
and factories. Fluke Ti25 thermal camera with fusion technology
was used to capture the images. The thermal imager consisted of

a 160 120 focal plane array, uncooled microbolometer detector

and operated in the infrared spectral band of 7.514 m. The thermal lens capture images of 320 240 pixels while the ordinary lens
produced 640 480 pixels (visual images). All the infrared images
of electrical equipments were collected at the main switchboard
which is supplying the electricity to an ofce building. For capturing the image, the thermal imager orientation is directly facing to
the target electrical equipment in order to get an accurate measurement. The distance between the target electrical equipment and
the thermal imager is in the range of 0.51.0 m. Emissivity measures how well the surface emits energy. In this study, the thermal
picture of electrical equipments was captured from main switch
boards in buildings. The metal surface of equipments was painted
or oxidized and also some of them were covered by high-quality
electrical tape. The standard emissivity of most organic materials and painted or oxidized surfaces is 0.95 [21]. Most low voltage
and many medium voltage switchgear and components have high
emissivity materials near the connection points. From molded case
breakers to cable insulation, an emissivity of 0.95 should perform
better [22]. It was noted that the ambient temperature around the
equipments is between 30 and 33 C during the inspection. A total
of 500 infrared images with different electrical equipments were
captured.
Conventionally, the condition of electrical equipment is evaluated by comparing the temperature value between abnormal
(hot) and reference (normal) components. This technique is known
as qualitative measurement which is analysed by measuring the
temperature difference, T between normal and abnormal components. Usually, the temperature of normal (reference) component is
assigned as minimum temperature and the temperature of abnormal (hot) component is assigned as maximum temperature. Several
standards for measuring T are found such as InterNational Electrical Testing Association (NETA) [23], American Society for Testing &
Materials (ASTM) E [24] and National Fire Protection Association
(NFPA) NFPA 70-B [25]. In the present study, based on the experience from inspections results, the conditions of equipments are
classied instead of using the available standards. For an accurate
temperature measurement, some important parameters related to
environmental effect, target equipment conditions as well as the
technique used for capturing the image have to be considered [26].
In this research, the conditions of equipments are classied into
three classes which was normal, warning and critical condition.
This specication is illustrated in Table 1 with their corresponding
recommended audit actions. Based on our survey on 500 electrical equipments 125 equipments are classied as critical, 193
equipments as warning and 182 equipments as normal condition.
This results show that 25% of the equipment had to be repair
immediately and 38% indicate probable deciency. A huge saving of energy is expected if the equipments are being replaced
immediately. For example, thermal insulation survey of a 460 MW
thermal power station with four units in India shows that about
1.02 million kcal/h energy losses was occurring due to bare surfaces, inadequate/damaged insulation or open cladding condition
in all four units. Further analysis shows that if these faulty insulated areas are attended there would be energy saving of around
0.774 million kcal/h [27]. All these equipments were evaluated
manually using infrared image analysis software. Some examples of
the conditions are depicted in Fig. 1, where the abnormal condition
is clearly shown by the red colour.

A.S. Nazmul Huda et al. / Energy and Buildings 55 (2012) 585591

587

Fig. 1. Thermal image of (a) abnormal condition, (b) warning condition, and (c) normal condition of electrical equipments. (For interpretation of the references to color in
text, the reader is referred to the web version of the article.)

Fig. 2. (a) Original infrared image of a three-phase cable, and (b) greyscale image with reference and hot region selections.

3. Infrared image processing and automatic features

extraction
Infrared images are converted into greyscale image where the
black pixels and white pixels imply the lowest and highest pixel
intensity region respectively. Therefore, hotspot of the electrical
equipment with respect to the reference spot carries more bright
pixels. Fig. 2 illustrates an example of infrared image and its corresponding greyscale image of a three-phase cable connection. The
suspected component (hot) is located at most right while reference
component is at the left. The equal size ROIs are manually selected
for both reference and hot components. Then rst and second order
statistical features are extracted from these ROIs. The features are
then used as the inputs of MLP networks for automatic condition
classication of electrical equipments.

the histogram of the pixel intensity and l is the possible intensity

level of the image. Average intensity is the average pixels value
which determines the brightness or darkness of the gray level
image. Skewness determines the asymmetrical property of the hisTable 2
First order statistical features.
Features

Formula

Maximum intensity

max

l1


Average intensity, 

1
nx ny

ip(i)

i=0
l1

Variance,  2

1
nx ny

3.1. Histogram based rst order statistical features

Histogram based rst order statistical features are very common
to describe the asymmetry in infrared images. Image histogram
is dened as the graphical representation of number of pixel values at different intensities level of that image [18]. The histogram
based rst order statistical features used in this study are average
grey level intensity, variance, skewness, kurtosis, entropy, standard
deviation and maximum grey intensity. The formulas of these features are summarized in Table 2:
In Table 2, nx and ny denote the number of pixels columns (width
of ROI) and number of pixels rows (height of ROI), respectively. i
is the number of distinct grey level in the quantized image. p(i) is

ip(i)

i=0
l1

(i ) p(i)

i=0
l1

Skewness

1
 3 nx ny

Kurtosis

1
 4 nx ny

(i ) p(i) 3

i=0

l1


(i ) p(i)

i=0

l1


Entropy

nx1ny

Standard deviation

 i=0

l1
 
2
 nx1ny
(i ) p(i)

i log(i)

i=0

588

A.S. Nazmul Huda et al. / Energy and Buildings 55 (2012) 585591

togram with respect to average intensity. If skewness is negative,

the grey pixel intensity of entire image is less than the average
grey level pixel intensity. The zero value of skewness means the
equal number of grey pixel intensity with respect to mean intensity.
The kurtosis measures variation of intensity distribution peakness
or atness with respect to the normal distribution (kurtosis = 3).
Entropy measures the disorderliness of the image and maximum
grey intensity denes the maximum pixel intensity of the greyscale
image. Variance determines the dispersion of grey level pixels from
the mean and standard deviation of pixel intensities is as like as
variance, but different in value.
The average grey intensity, entropy, maximum grey intensity,
kurtosis, skewness, standard deviation and variance for the region
of hot component are denoted as x1 , x2 , x3 , x4 , x5 , x6 and x7 ,
respectively. Similarly, the similar parameters also applied for the
reference component which can be represented as h1 , h2 , h3 , h4 , h5 ,
h6 and h7 . Besides, mutual parameters i.e. the differences between
hot and reference parameters are calculated namely as d11 (h1 x1 ),
d12 (h2 x2 ), d13 (h3 x3 ), d14 (h4 x4 ), d15 (h5 x5 ), d16 (h5 x5 ), and
d17 (h7 x7 ).
3.2. Statistical features using GLCM
The basic difference between rst and second order statistical
approach is that the rst order statistics estimate only the properties of individual pixel values, while second order statistics estimate
spatial relationships between pixel grey levels of the image occurring at specic locations relative to each other [28]. A GLCM is a
square having number of rows and columns equal to the number
of grey levels in the image that is formed on the probabilities that
pairs of pixels, separated by a certain pixel distance and a specic
direction [29].
In this research, a distance of 1 pixel is considered i.e.
neighbouring pixels in the pair. Mostly small distances produce
the best results [30], since they are appropriate for textures that are
ne, as well as for those that are coarse. For a given distance d, there
are four angular co-occurrence matrices along four directions: 0
(horizontal), 45 (diagonal), 90 (vertical) and 135 (anti-diagonal).
Selection of appropriate direction is very difcult. The most common choice for the direction between pixels is 0. In this work,
co-occurrence matrix along 0 is calculated. This means that the
pixels in the pair are located horizontally with respect to each other.
Thus at rst ROIs are selected and GLCM features are evaluated
from both the hottest and reference components. Haralick et al.
suggested that 14 features can be derived from GLCM [31].The statistical features using GLCM utilized in this study are homogeneity,
energy, entropy and contrast. The formulas of these features are
listed in Table 3.
Where, i denotes the grey level of the reference pixel, j is the
grey level of neighbouring pixel and p(i,j) is the normalised GLCM.

Table 3
GLCM features.
Features

Formula
l1
l1



Homogeneity
i=0
l1

p(i,j)
1+|ij|

j=0
l1


Energy

p2 (i, j)

i=0
j=0
l1
l1



Entropy

i=0
l1

p(i, j)log2 p(i, j)

j=0
l

l
 

n2

Contrast
n=0

i=1

p(i, j), |i j| = n

j=1

Homogeneity measures the closeness of the distribution of elements in the co-occurrence matrix with respect to its diagonal.
Energy is the sum of square elements in the co-occurrence matrix.
Contrast is a measure of the intensity contrast between a pixel and
its neighbour over the whole image and entropy used to calculate statistical randomness. The homogeneity, energy, entropy and
contrast for the ROI of the hot component are denoted as p1 , p2 , p3 ,
p4 , respectively. Similarly for the reference component: g1 , g2 , g3 ,
g4 are represented for homogeneity, energy, entropy and contrast
of its region. The mutual parameters are calculated by taking the
difference between hot and reference parameters. These parameters
are given as d21 (p1 g1 ), d22 (p2 g2 ), d23 (p3 g3 ), and d24 (p4 g4 ).
But the conditions of equipments are mainly inuenced by the
characteristics of the hot component and difference of parameters
between hot and reference. Therefore, total 22 features are primarily selected neglecting 11 features of reference parameters which
are highlighted in bold as shown in Table 4.
4. Suitable features selection
Features selection is used for reducing data dimensionality. All
extracted features from 500 infrared images of electrical equipments in building are tabulated into three categories. Principle
component analysis (PCA) tool was employed for features selection
from the above mentioned of 22 features. The principle components are calculated using correlation matrix. The result of PCA is
shown in scree plot which is used to justify relative magnitude of
eigenvalues. Fig. 3 shows the scree plot for 22 features of all the
data set. The results of scree plot and eigenvalues as illustrated
in Table 5. Eigen analysis of the correlation matrix for electrical
equipments conditions classication data Table 5. It is found that
the most appropriate number of principle component is four for the
rst principle component (PC1) until fourth principle component

Table 4
Features extracted from both hot and reference region.
Feature type

List of parameters

Reference parameter

Hot parameter

Difference between
hot and reference

Histogram based 1st order features

Average grey intensity

Entropy
Maximum grey intensity
Kurtosis
Skewness
Standard deviation
Variance

h1
h2
h3
h4
h5
h6
h7

x1
x2
x3
x4
x5
x6
x7

d11
d12
d13
d14
d15
d16
d17

Statistical features using GLCM

Homogeneity
Entropy
Energy
Contrast

g1
g2
g3
g4

p1
p2
p3
p4

d21
d22
d23
d24

A.S. Nazmul Huda et al. / Energy and Buildings 55 (2012) 585591

589

Table 6
Eigenvectors of PC1PC4 for electrical equipments conditions classication.

Fig. 3. Scree plots of conditions classication data.

(PC4). PC1 has variance (eigenvalue) of 6.8786 and accounts for

31.3% of the total variance. On the other hand, PC2, PC3 and PC4
have the eigenvalues of 4.9777, 2.7349 and 1.9782 producing their
respected total variance of 22.6%, 12.4% and 9.0%. Therefore, the
rst four principle components give 75.3% of the total variance. The
remaining principal components produce very small variance and
thus these are assumed as unimportant. The eigenvectors of the
four principle components (PC1PC4) are shown in Table 6. The
variables that have strongly correlated with PC1PC4 are tabulated
in Table 7.

Variable

PC1

PC2

PC3

PC4

x1
x2
x3
x4
x5
x6
x7
d11
d12
d13
d14
d15
d16
d17
p1
p2
p3
p4
d21
d22
d23
d24

0.229
0.132
0.084
0.222
0.230
0.112
0.113
0.124
0.074
0.016
0.062
0.008
0.058
0.079
0.320
0.342
0.355
0.237
0.271
0.324
0.343
0.251

0.054
0.137
0.182
0.175
0.178
0.317
0.314
0.244
0.191
0.324
0.151
0.151
0.405
0.403
0.055
0.013
0.021
0.080
0.158
0.174
0.157
0.143

0.313
0.259
0.299
0.147
0.208
0.275
0.272
0.231
0.352
0.013
0.342
0.380
0.028
0.124
0.023
0.104
0.030
0.198
0.040
0.076
0.015
0.133

0.077
0.020
0.071
0.315
0.258
0.260
0.244
0.339
0.282
0.318
0.099
0.125
0.119
0.002
0.064
0.214
0.187
0.369
0.269
0.012
0.034
0.260

Table 7
Variables that have strong relationship for PC1PC4.
PC1

PC2

PC3

PC4

p1
p2
p3
d22
d23

x6
d13
d16
d17

d12
d14
d15

x4
d11
p4

5. Classication using MLP network

After choosing the best features, the next step is to classify
the conditions of equipments using MLP neural network. In MLP
architecture, the network contains three layers i.e. input layer, hidden layer and output layer as illustrated in Fig. 4 [32]. The rst
and the last layers are called as input and output layers, respectively. In the input layer, the number of nodes corresponds to the
number of input features and the number of nodes in the output
layer corresponds to the number of target classes. Neurons are the
Table 5
Eigen analysis of the correlation matrix for electrical equipments conditions classication data.
Principle component

Eigen value

Proportion

Cumulative

PC1
PC2
PC3
PC4
PC5
PC6
PC7
PC8
PC9
PC10
PC11
PC12
PC13
PC14
PC15
PC16
PC17
PC18
PC19
PC20
PC21
PC22

6.8786
4.9777
2.7349
1.9782
1.2864
0.9567
0.7587
0.6125
0.4627
0.3615
0.2719
0.2160
0.1319
0.1149
0.0958
0.0761
0.0350
0.0168
0.0155
0.0121
0.0033
0.0027

0.313
0.226
0.124
0.090
0.058
0.043
0.034
0.028
0.021
0.016
0.012
0.010
0.006
0.005
0.004
0.003
0.002
0.001
0.001
0.001
0.000
0.000

0.313
0.539
0.663
0.753
0.812
0.855
0.890
0.917
0.938
0.955
0.967
0.977
0.983
0.988
0.993
0.996
0.998
0.998
0.999
1.000
1.000
1.000

information-processing unit that is the fundamental to the operation of a neural network.

Consider, a standard MLP network with inputs z1 , z2 ,. . ., zni ,
predicted outputs y1 , y2 ,. . ., ym and nh hidden nodes. Connection
weights between input and hidden layer are denoted as wij1 while
2 denotes the connection weights between hidden and output
wjk
layer. The network is fully connected in the sense that every neuron in each layer of the network is connected to every other neuron
in the adjacent forward layer by the weighted connection. Also bj
denotes the thresholds in the hidden nodes. Then the predicted
output of the jth neurons of the kth node in the output layer of the

Fig. 4. Standard MLP architecture.

590

A.S. Nazmul Huda et al. / Energy and Buildings 55 (2012) 585591

Table 8
Input and output parameters for MLP neural network.
Number of data
Training data
Testing data
Input features, zn = 15

Table 9
Classication performance for testing data set.

500
300
200
p1 , p2 , p3 ,d22 , d23 , x6 ,d13 , d16 , d17 , d12 ,
d14 , d15 , x4 , d11 , p4
Critical, warning and normal condition

Output classes, ym = 3

MLP network can be expressed as:

y k (t) =

nh


2
wjk
F

j=1

 n


wij1 zi (t) + bj

for 1 k < m and 1 j < nh

i=1

(1)
In this experiment, fteen input nodes and three output nodes
were being used. The target outputs are classied into three categories i.e. normal, critical, and warning condition. The F() is the
sigmoid activation function that is commonly used in the MLP net2 and
work which is also used in this study. The weights wij1 , wjk
threshold bj are unidentied and should be chosen for reducing
the error in prediction. The prediction error is dened as follows:
k (t) = yk (t) y k (t)

(2)

yk (t) are the actual (i.e. target) and predicted outwhere yk (t) and 
put, respectively.
6. Classication performance analysis
The entire data set containing 500 images are divided into training data and testing data set. In this research, 60% of the data are
selected randomly as for data training and remaining 40% of the
data are used for testing. Table 8 summarizes the input and output
parameters for classication.
MLP network has been trained and compared by using twelve
different training algorithms namely LM, BFG, RP, SCG, CGB, CGF,
CGP, OSS, BR, GD, GDX and GDM to nd out the best training algorithm for classication. The numbers of hidden neuron are varied
from 1 to 50 for the networks to nd the best performance accuracy.
The number of epochs is kept at constant 2000 for all networks. For
comparing the performance of different classiers, the percentage
of true classication for the normal condition (Accnormal ), warning condition (Accwarning ), critical condition (Acccritical ) and overall
(Accoverall ) are evaluated, which can be dened as follows:

Training
algorithm

No. of
hidden
nodes

Accoverall
(%)

Accnormal
(%)

Accwarning
(%)

Acccritical
(%)

LM
BFG
RP
SCG
CGB
CGF
CGP
OSS
BR
GD
GDX
GDM

9
19
21
14
29
39
29
10
10
34
32
21

78.5
76.5
77.0
78.0
76.5
77.5
78.0
78.0
76.0
58.5
74.5
63.0

90.7
81.8
83.6
85.5
73.5
89.4
83.8
89.4
75.0
71.2
90.0
76.4

59.7
72.6
68.5
70.3
79.5
67.1
74.1
66.3
75.3
50.6
63.9
45.9

86.8
74.0
84.1
78.0
77.1
79.6
76.5
82.4
78.7
52.0
70.2
68.5

Table 10
Classication using discriminant analysis.
True group
Put into group
Critical
Warning
Normal
Total N
N correct
Accuracy (%)

Critical
93
30
2
125
93
74.4

total number of equipments condition classied correctly

100%
total number of equipments in dataset

(3)

Accnormal =

total number of normal equipment condition classied correctly

100%
total number of normal equipments condition in dataset

(4)

Acccritical =

Normal

Overall

36
117
40
193
117
60.6

1
33
148
182
148
81.3

130
180
190
500
358
71.6

58.5%, respectively. MLP neural network trained with LM algorithm

was able to achieve the highest classication accuracy of critical
condition (86.8%). Thus, this network has maximum capacity to
classify the critical condition equipments correctly with respect to
total number of critical condition equipments in the testing data
set. Similarly, MLP network trained with CGB algorithm was able
to achieve the highest classication accuracy of warning condition (79.5%). Therefore, MLP network trained with CGB algorithm
has highest capacity to classify the warning condition equipments
correctly with respect to the total number of warning condition
equipments in the testing data set. Also MLP network trained with
LM algorithm gives the maximum normal condition classication
accuracy (90.7%). On the other hand, MLP network trained with GD
algorithm gives the worst testing overall accuracy 58.5% providing
the lowest percentage of classication accuracy for all conditions

Accoverall =

Accwarning =

Warning

total number of warning equipments condition classied correctly

100%
total number of warning equipments condition in dataset

total number of critical equipments condition classied correctly

100%
total number of critical equipments condition in dataset

(5)
(6)

7. Result and discussion

Table 9 shows the testing performance of MLP neural network
for twelve different training algorithms. The results of training data
set are not mentioned in this section, as the data only used to learn
the network. MLP network trained with LM algorithm gives the
best testing overall accuracy 78.5%. Then followed by MLP network trained with SCG, OSS, CGP, CGF, RP, BFG, CGB, BR, GDX,
GDM and GD algorithms give the testing overall accuracies of 78.0%,
78.0%, 78.0%, 77.5%, 77.0%, 76.5%, 76.5.0%, 76.0%, 74.5%, 63.0% and

(normal: 71.2%; warning: 50.6% and critical: 52.0%). The results also
show that the MLP network trained with LM need lower number of
hidden nodes which followed by OSS, BR, SCG, BFG, RP, GDM, CGP,
CGB, GDX, GD and CGF training algorithms.
Table 10 shows the classication performance of three conditions of electrical equipments using discriminant analysis (DA)
classier. Total numbers of data and input features are 500 and
15 respectively. The classication accuracy for critical condition is
74.4% (93 critical conditions correctly classied out of 125), 60.6%

A.S. Nazmul Huda et al. / Energy and Buildings 55 (2012) 585591

Table 11
Comparison of performance between MLP and DA classier.
Classier

MLP trained with LM

DA

Accoverall (%)
Accnormal (%)
Accwarning (%)
Acccritical (%)

86.8
90.7
59.7
78.5

71.6
81.3
60.6
74.4

for warning condition (117 warning conditions correctly classied

out of 193) and 81.3% for normal condition (148 normal conditions
correctly classied out of 182). Overall accuracy is dened as:
Accoverall =

number of correct classied data

total number of data

100%

(7)

Therefore, overall accuracy produced by DA classier is about

71.6%. As for comparison, the classication accuracy between MLP
neural network and DA classier are shown in Table 11. MLP network trained with LM gives the best testing performance with the
total accuracy up to 78.5% (normal: 90.7%; warning: 59.7% and Critical: 86.8%) compared to DA which can give only 74.4% accuracy.
8. Conclusion
In this study, an advanced method for electrical equipment protection in buildings by inspecting thermal conditions of electrical
installations is proposed. The system employs the multilayered perceptron neural network trained using different training
algorithms. The regions of interest for faulty equipment and reference equipment were manually segmented. Histogram based
rst order features and grey level co-occurrence matrices features were extracted from regions of interest. Using principle
component analysis, 15 features are selected as the inputs of
neural network. Multilayered perceptron network trained with
LevenbergMarquardt produced the highest testing accuracy of
78.5%. Additionally, the results of multilayered perceptron network trained with LevenbergMarquardt also compares with the
performance of discriminant analysis classier. The difference of
performance between multilayered perceptron network and discriminant analysis classier proves that multilayered perceptron
network trained with LevenbergMarquardt algorithm as the best
classier. As a scope for future work, automatic diagnosis system
with suitable feature extraction can be considered for improving
the performance of the system.
Acknowledgements
The authors gratefully acknowledge the contribution of Kamarul
Hawari bin Ghazali, Universiti Malaysia Pahang for helping to
capture thermal images. The work is supported by Fundamental
Research Grant Scheme (203/PELECT/6071211), Universiti Sains
Malaysia.
References
[1] Statistics Fire Causes Building Fire Investigation the Fire Department
and Rescue Malaysia, Year 2012, Nationwide, From January to June.
http://www.bomba.gov.my/leupload/statistik/ST-PGK%20PK01(2012).pdf,
2012 (accessed 01.08.12).
[2] V. Babrauskas, How do electrical wiring faults lead to structure ignitions, in:
7th international Fire & Materials conference, San Francisco, 2001, pp. 3950.

591

[3] M. Braunovic, N.K. Myshkin, V.V. Konchits, Electrical Contact Fundamentals,

Applications and Technology, CRC Press Taylor and Francis Group, 2007, Ch
11(Monitoring Technologies).
[4] S. Martin, I. Canas, I. Gonzalez, Thermographic survey of two rural buildings in
Spain, Energy and Buildings 34 (2004) 515523.
[5] E. Grinzato, V. Vavilov, T. Kauppinen, Quantitative infrared thermography in
buildings, Energy and Buildings 29 (1998) 19.
[6] N.P. Avdelidis, A. Moropoulou, Emissivity considerations in buildings thermography, Energy and Buildings 35 (2003) 663667.
[7] E. Barreira, V.P. Freitas, Evaluation of building materials using infrared thermography, Construction and Building Materials 21 (2007) 218224.
[8] F. Asdrubali, G. Baldinelli, F. Bianchi, A quantitative methodology to evaluate
thermal bridges in buildings, Applied Energy, in press.
[9] C.A. Balaras, A.A. Argiriou, Infrared thermography for building diagnostics,
Energy and Buildings 34 (2002) 171183.
[10] M.A. Kregg, Benets of using infrared thermography in utility substations,
Proceedings of SPIE, 2004, pp. 249257.
[11] F. Lizak, M. Kolcun, Improving reliability and decreasing losses of electrical
system with infrared thermography, Acta Electritechnica et Informatica 81
(2008) 6063.
[12] R.A. Epperly, G.E. Heberlein, L.G. Eads, A tool for reliability and safety: predict and prevent equipment failures with thermography, IEEE Conference on
Petroleum and Chemical Industry (1997) 5968.
[13] Y.C. Chou, L. Yao, Automatic diagnosis system of electrical equipment using
infrared thermography, International Conference of Soft Computing And Pattern Recognition (2009) 155160.
[14] M.S. Jadin, S. Kabir, S. Taib, Thermal imaging for qualitative-based measurements of thermal anomalies in electrical components, in: IEEE Conference on
ECPC, Saudi Arab, 2011.
[15] C.A.L. Almeida, A.P. Braga, S. Nascimento, V. Paiva, H.J.A. Martins, R. Torres, W.M. Caminhas, Intelligent thermographic diagnostic applied to surge
arresters: a new approach, IEEE Transactions on Power Delivery 24 (2) (2009)
751757.
[16] M.A. Shai, N. Hamzah, Internal fault classication using articial neural network, in: Proceedings of 4th International Power Engineering and Optimization
Conference, 2010, pp. 352357.
[17] A. Rahmani, J. Haddadnia, O. Seryasat, Intelligent fault detection of electrical
equipment in ground substations using thermo vision technique, in: Proceddings of 2nd International Asia Conference on Mechanical And Electronics
Engineering, vol. 2, Kyoto, 2010, pp. V2V150.
[18] B. Wiecek, M. Strzelecki, T. Jakubowska, C. Drews-Peszynski, Advanced Thermal
Image Processing, Technical University of Lodz, Taylor and Francis Group, 2006.
[19] I.T. Jolliffe, Principle Component Analysis, Springer-Verlag, 1986, 487.
[20] K.K. Aggarwal, Y. Singh, P. Chandra, M. Puri, Bayesian regularization in a neural network model to estimate lines of code using function points, Journal of
Computer Sciences 1 (4) (2005) 505509.
[21] Why Thermography is Good for Your Business. http://www.newark.com/pdfs/
techarticles/Why Thermography is Good for Business.pdf, 2012 (accessed
01.08.12).
[22] Protecting Buildings Against Electrical System Failures. http://www.
plantservices.com/wp downloads/pdf/PS090326 Flir.pdf, 2012 (accessed
02.08.12).
[23] Standard for Infrared Inspection of Electrical Systems & Rotating Equipment, Infraspection Institute, 2008. http://www.armco-inspections.com/
les/ir/Electrical%20Rotating%20Std.pdf, 2011 (accessed 11.06.11).
[24] ASTM, ASTM E 1934: Standard Guide for Examining Electrical and Mechanical
Equipment with Infrared Thermography, ASTM International, West Conshohocken, PA, 2005.
[25] NFPA, NFPA 70B: Recommended Practice for Electrical Equipment Maintenance, National Fire Protection Association, Quincy, MA, 2006.
[26] M.S. Jadin, S. Taib, Recent progress in diagnosing the reliability of electrical
equipment by using infrared thermography, Infrared Physics & Technology 55
(2012) 236245.
[27] S.P. Garnaik, Infrared Thermography: A Versatile Technology for Condition
Monitoring and Energy Conservation, 2011, http://www.reliabilityweb.com/
art05/Infrared thermography a versatile technology.pdf (accessed 01.07.11).
[28] J.R. Jensen, Remote Sensing of the Environment: An Earth Resource perspective,
Prentice-Hall, NJ, 2000, p. 336.
[29] C.M. Miquel, Texture discriptors applied to digital mammography. M.Sc. Thesis,
University of Girona, 2009.
[30] V. Karathanassi, C.H. Iossidis, D. Rokos, A texture-based classication method
for classifying built areas according to their density, International Journal of
Remote Sensing 21 (9) (2000) 18071823.
[31] R.M. Haralick, K. Shanmugan, I. Dinstein, Texture features for image classication, IEEE Transactions on Systems Man and Cybernetics 8 (1973) 610621.
[32] S. Haykin, Neural Networks: A Comprehensive Foundation, 2nd ed., Prentice
Hall, NJ, 1988.