You are on page 1of 7

Electrical Power and Energy Systems 63 (2014) 753–759

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

journal homepage:

Design of early fault detection technique for electrical assets using

infrared thermograms
Zainul Abdin Jaffery, Ashwani Kumar Dubey ⇑
Department of Electrical Engineering, Faculty of Engineering and Technology, Jamia Millia Islamia, New Delhi, India

a r t i c l e i n f o a b s t r a c t

Article history: Infrared thermography is a non destructive temperature measuring technique, widely used in agriculture,
Received 4 September 2013 process industries, manufacturing, pharmaceutical industries, and construction industries to detect
Received in revised form 23 May 2014 anomalies, to predict possible faults, or to check the quality of the object. In this paper, two systems, real
Accepted 16 June 2014
time and off line, are proposed to monitor the temperature variations and analyze hot regions in the elec-
Available online 16 July 2014
trical assets using infrared thermograms. This novel technique helps to prevent the electrical assets
before any catastrophe would happen in the future. The color based segmentation technique is used to
blotch hot regions in the thermograms of electrical systems. A redness area based algorithm is also pro-
posed to analyze the hot regions and also to estimate rate of change of hotness in electrical assets for
Power transformer early detection and prediction of faults.
Segmentation Ó 2014 Elsevier Ltd. All rights reserved.

Introduction transformers, induction motors, and synchronous motors.

Whereas, off line monitoring includes: inspection and analysis of
Recently, the infrared (IR) thermograph technology has gained electrical contacts, excessive current drawn, fuse cabinets, imbal-
more recognition and acceptance due to its non-contact and non- ance loads, inductive heating, corrosion, defective components,
destructive features of inspection. It is a fast and reliable inspection feeder poles, capacitor banks, generator controls and transfer
system that operates without interrupting the running operation of switches, main incoming line electrical panels, motor control cen-
power system. In IR thermograph based technique, fault diagnosis ters, cables and bus bars. The off line monitoring of the electrical
is performed through the analysis of thermal image captured by systems can be done through regular thermographic inspections
infrared camera. It is implicit that the life of electrical equipment under operational conditions to identify problem early and avoid
is radically reduced as temperature rises. Thermogram based tem- expensive equipment failure.
perature measurement technique offers many advantages such as
prompt response times, ample temperature ranges, highly reliable, Related works
harmless, high spatial resolution, and very lucrative approach for
the monitoring of electrical power systems. As an upshot of liberalization, investments in new electrical
The thermogram is solely based on the heat distribution in the power system equipment have radically declined over the past
system. The image segmentation from the thermograms becomes 20 years. Many transformers are working well beyond their antici-
complex due to its low intensity contrast and over-centralized pated life and are operating under escalating stress. As load is grow-
intensity distribution. The extraction of the hot region within a ing, new generation and economically aggravated transmission
thermogram is an exigent task, particularly if the image has low flows push equipment to work beyond the specification. As a conse-
signal to noise ratio (SNR) and complex background. quence, new techniques must be explored to allow electrical equip-
In this work, the non invasive infrared image based visual mon- ment to better fit under such circumstances and also be
itoring and controlling systems are developed for both, real time economically acceptable and reliable [1,2]. The major flow of cost
and off line applications. The real time application may include involved in the monitoring of power grid to ensure quality and
the continuous monitoring of electrical machines such as power uninterruptable supply of electricity [3]. The equipment on which
whole grid system depends is transformer and the monitoring of
⇑ Corresponding author. Tel.: +91 9953289235. transformer in efficient way is still a big task. In most of the existing
E-mail addresses: (Z.A. Jaffery), systems, monitoring of transformer is done through the dissolved
(A.K. Dubey). gas analysis (DGA) technique [4,5]. But this technique requires an
0142-0615/Ó 2014 Elsevier Ltd. All rights reserved.
754 Z.A. Jaffery, A.K. Dubey / Electrical Power and Energy Systems 63 (2014) 753–759

exhaustive and costly procedure to detect faults. Few mathematical marching, and generation of control signals according to character-
models [6] and fuzzy rule [7] based techniques are also proposed to istics of image.
identify the faults, but, these techniques are also requires some
measured data to interpret and predict. Therefore, the non invasive Thermogram image segmentation
temperature monitoring of transformers [8] and electrical assets
with less complexity and high reliability is a prime requisite with In NIOLVIS the thresholding with smoothing technique is
existing system to overcome the maintenance cost and revenue loss applied for segmentation [19–26] of surface of desire from thermo-
during break down of the system. The infrared (IR) imaging tech- grams to explore the regions with abnormalities in the electrical
nique makes non invasive type monitoring systems more reliable systems. The mathematical representation of thresholding is given
for prophecy of temperature of electrical assets [9,10]. The IR tech- by (1):
niques are also being used to monitor house hold electrical equip- 
ment and wirings [11], determination of hydro generator 1; if Tði; jÞ > q
Lði; jÞ ¼ ð1Þ
efficiency [12], real-time thermal monitoring of induction 0; if Tði; jÞ < q
machines [13], measurement of excitation winding temperature
where T(i, j) and L(i, j) represents original and thresholded images
in synchronous generator [14], and real time high temperature
respectively and q is the specified threshold value. This segmenta-
measurement in control industry [15]. The design of visual inspec-
tion results the image into binary form, where, 1 represents the
tion system (VIS) [16–20] based on charge-coupled device (CCD)
desired color region and 0 the undesired color region. Otsu method
and complementary metal oxide semiconductor (CMOS) cameras
[27] is used for automatic selection of threshold value for effective
are already into existence for the monitoring of different processes,
segmentation of infrared image colors because this method does
objects, sorting, and quality check but designing VIS based on IR
not depend on modeling the probability density functions and
camera are very less.
assumes a bimodal distribution of gray-level values. Also, in this
In most of the papers discussed above, only invasive or pointed
method, only zeroth and first order cumulative moments of the gray
temperature is observed or measured and no non-invasive moni-
level histograms are used and integration based optimal threshold
toring and controlling systems were proposed or developed. But,
is selected. The optimal threshold value is estimated using (2) as
in our work, two IR imaging based visual monitoring and control-
ling systems: Non Invasive Off Line Visual Inspection System
( )
(NIOLVIS) and Non Invasive Real Time Visual Monitoring System PðqÞ½1  PðqÞ½l1 ðqÞ  l0 ðqÞ2
(NIRTVMS), are proposed for off line and real time applications qOT ¼ arg max ð2Þ
PðqÞr21 ðqÞ þ ½1  PðqÞr20 ðqÞ
respectively. The proposed systems have capability to estimate
area of the hotspots in the electrical assets and take decisions where qOT = optimal threshold value, l1 = mean values of desired
accordingly to keep other part of electrical systems healthy. The foreground image, l0 = mean values of undesired background
proposed algorithms are simulated for a large number of thermo- image, r1 = standard deviation of desired foreground image
gram images of electrical assets and their performance are ana- r0 = standard deviation of undesired background image. Also, PðqÞ
lyzed using statistical and geometrical features. The simulation is used in (2) is calculated by (3) and (4).
done on MATLABR2012a.
Cumulative probability PðqÞ ¼ pðkÞ ð3Þ
Non Invasive Off Line Visual Inspection System (NIOLVIS) k¼0

The process flow block diagram of NIOLVIS is shown in Fig. 1. fk

pðkÞ ¼ ð4Þ
This NIOLVIS includes an infrared image acquisition system, n
threshold and color based segmentation, feature extraction and
where pðkÞ = probability distribution of image histogram,
fk = frequency of gray level k, n = total number of pixel in the image.
In NIOLVIS, RGB color vectors [28] are used to get desired color
range by estimating average from the RGB color space and the
objective is to classify each RGB pixel in a given image having a
color in the specified range.
Consider the average color as RGB vector r and q as any point in
RGB space. The q will said to be analogous to r, if, the distance
between them is less than a specified threshold value q. This sim-
ilarity measure is estimated using Euclidean distance between q
and r as given in (5).
qðq; rÞ ¼ jjq  rjj ¼ ½ðq  rÞT ðq  rÞ
¼ ½ðqR  rR Þ2 þ ðqG  r G Þ2 þ ðqB  rB Þ2  ð5Þ

where subscript R, G, and B denotes the RGB components of vectors

q and r.
The smoothing [28,29] filter is used to remove noise and arti-
facts from an image prior to object extraction. This bridges the
gap between curves and lines [30–32]. Smoothing filter is the aver-
age of the pixels contained in the neighborhood of the filter mask
given by (6). As the mask is slid across the image to be smoothed,
each pixel is replaced by the average of the pixels in the neighbor-
hood defined by the mask. Consider, Pij denotes the set of coordi-
Fig. 1. Process flow block diagram of NIOLVIS. nates defining a neighborhood centered at (i, j) in an RGB color
Z.A. Jaffery, A.K. Dubey / Electrical Power and Energy Systems 63 (2014) 753–759 755

image. The average of the RGB component vectors in the neighbor-

hood is:
 ði; jÞ ¼
a aðl; mÞ ð6Þ
K ðl;mÞ2P ij

Feature calculation

Statistical features
The statistical features are used to detect the presence of object
of desire within the thermogram. The most prominent statistical
features are: variance (r2), arithmetic mean (l), and standard devi-
ation (r). These features are estimated through (7)–(9) as follows:
X Px;y
Mean of image pixel values ðlÞ ¼ ð7Þ
X Px;y  Px;y
Standard deviation of image ðrÞ ¼  ðl2 Þ ð8Þ

Variance of image ¼ r2 ð9Þ

where Px,y represents the pixel values at x, y location in the seg-

mented image.

Geometrical features
Geometrical features of a thermographic image are a set of geo-
metrical elements such as curves, lines, surfaces, points, orienta-
tion, and primitive features like blobs, corners, edges, and ridges
[27,33–35]. In this paper, number of blobs and blob area of red
color region is estimated. The blob area represents the size of
image foreground [20,35–37].

Implementation and results

In this work, an infrared thermographic image of a fuse cabinet

is taken for fault diagnosis and analysis using NIOLVIS for off-line
Fig. 3. Histogram of RGB bands.
monitoring application. The original infrared thermogram of fuse
cabinet is shown in Fig. 2 and the histograms of RGB (red, green,
and blue) bands available in thermogram are shown in Fig. 3.
segmentation process e.g. thresholded image, region filling, border
The features like: area, location, mean and number of hot blobs
smoothing, and masking of the desired region. The final segmented
in thermogram are estimated after segmentation of hot regions
image is shown in Fig. 5. In this figure, five blobs are appearing. The
(red regions) from the thermogram. Fig. 4 is showing the steps of
number of blobs in the segmented region depends upon the thres-
holding value of filter. For the simulation, the threshold value is
empirically chosen as 50 pixels. This means, the region having pix-
els less than 50 are removed. This threshold value may be opti-
mized using any optimization technique. The fuzzy knowledge
base reasoning may also be used for threshold selection.
The features of segmented image shown in Fig. 5 are computed
and summarized in Table 1. This image has five blobs. The features
like: area of each blob in pixel and mean values of each RGB com-
ponents are analyzed. The NIOLVIS interprets the results according
to the type and specifications of the systems.

Non Invasive Real Time Visual Monitoring System (NIRTVMS)

for power transformer

The block diagram of non-invasive real time visual monitoring

system is shown in Fig. 6. The process flow diagram for NIRTVMS
is similar to that of NIOLVIS as given in Fig. 1. The major elements
of NIRTVMS are: an infrared image acquisition system, an image
segmentation technique, a segmented image analysis technique,
and generation of control signals. In the following sub sections,
core stages of NIRTVMS are explained. Which includes: IR image
Fig. 2. Original infrared image of fuse cabinet. segmentations, feature calculations, and feature matching. This
756 Z.A. Jaffery, A.K. Dubey / Electrical Power and Energy Systems 63 (2014) 753–759

Fig. 6. Block diagram of proposed NIRTVMS.

Start monitoring of power transformer

Initialize infrared camera to capture images

Segment the different colors from infrared image

Fig. 4. Intermediate stages of segmentation.

Whether all colors are

Calculate features of red colored blobs

Whether the area of RED Y

color region (Ai) is greater
than upper threshold limit
Generate tripping &
alarming signal

Whether the area of RED color Y

region (Ai) is greater than
medium threshold limit (ATM)?

Generate controlling &

alarming signal
Fig. 5. Final segmented image.

Table 1
Whether the area of RED Y
Features of red color region.
color region (Ai) is greater
than lower threshold limit
Blob number Area in pixels Mean R Mean G Mean B (ATL)?
1 197 215.40 96.64 31.21 Generate alarming
2 5326 228.46 85.43 46.81
3 474 234.11 60.70 37.23
4 67 244.27 129.63 3.57
5 70 216.53 59.29 11.79 System is healthy/
Keep monitoring

NIRTVMS have a facility to store frames in a queue, and retrieve the

frames from the queue for further processing [38]. Fig. 7. Flow chart of proposed NIRTVMS.

IR image segmentation Feature calculation

The segmentation technique used for NIRTVM is same as given The statistical and geometrical features are evaluated as given
‘Thermo image segmentation’. ‘Feature calculation’.
Z.A. Jaffery, A.K. Dubey / Electrical Power and Energy Systems 63 (2014) 753–759 757

Fig. 9. (a) and (b) Are showing filtering and thresholding steps for segmentation of
only red color regions. (For interpretation of the references to color in this figure
legend, the reader is referred to the web version of this article.)

Fig. 10. Final segmented image.

(image features of desired image), are used for similarity measure-

ment by (10):
z X

Fig. 8. (a) Original thermogram of power transformer with combined histogram of Si ðM Obs ; M Ref Þ ¼ TraceðABT BAT Þ ¼ cos2 /qr ð10Þ
RGB colors and masked red objects. (b–d) image contents along with histogram and biq¼1 r¼1
masking information for red, green, and blue bands respectively. (For interpretation
of the references to color in this figure legend, the reader is referred to the web where A, B are the matrices having first p principle components of
version of this article.) MObs, MRef and /qr is the angle between qth principal component of
MObs and rth principal component of MRef. The range of similarity
factor spans from 0 to z.
Feature matching
Proposed algorithm for temperature analysis in real time
Principal Component Analysis (PCA) method [20] is used to
match the image features in NIRTVMS. In this method, two matri- Let the total area of the object is A and the total area of red color
ces MObs (image features of captured real time image) and MRef regions (blobs) at any instant I is given by (11):
758 Z.A. Jaffery, A.K. Dubey / Electrical Power and Energy Systems 63 (2014) 753–759

Table 2 and decreases with increase in distance of camera from the surface
Geometrical features of red color region. of interest (SOI).
Sets Area Convex area Filled area The controlling and alarming [39] threshold values of area of
TR10 28,241 60,363 33,339
redness (AOR) may be taken as follows:
TR11 28,447 60,038 33,579 Upper Threshold Area (ATU) = 80% of A.
TR12 28,962 67,887 33,654 Medium Threshold Area (ATM) = 70% of A.
TR13 31,802 84,471 33,180 Lower Threshold Area (ATL) = 60% of A.
TR14 27,726 59,078 32,548
These threshold values can be optimized and varied according
TR15 27,516 60,596 32,081
TR16 28,864 62,734 29,371 to the core area of transformer and type of cooling systems.
TR17 27,463 61,361 32,195
TR18 28,135 59,640 32,927
TR19 28,322 66,495 32,865 Implementation and results
TR20 28,343 68,089 29,290
The flow chart for complete functional algorithm of NIRTVMS is
shown in Fig. 7. This flow chart includes: initialization of infrared
camera, segmentation, image feature calculation, feature matching,
Table 3 alarming, and the controlling actions. The four decision boxes in
Statistical features of red color region. flow chart shows the sharp reasoning and decision making capabil-
Sets Mean SD Solidity ity of the system. The simulation results of internal image process-
ing steps applied in NIRTVMS are shown in Fig. 8. In this figure,
TR10 0.2326 0.4225 0.4679
TR11 0.2349 0.4239 0.4738 Fig. 8(a) includes original thermogram of power transformer, com-
TR12 0.2381 0.4259 0.4266 bined histogram of RGB colors, and masked red objects. Fig. 8(b–d)
TR13 0.2551 0.4359 0.3765 have image contents along with histogram and masking informa-
TR14 0.2247 0.4174 0.4693
tion for red, green, and blue bands respectively.
TR15 0.2251 0.4176 0.4541
TR16 0.2247 0.4174 0.4601
Fig. 9 is showing the steps of segmentation of desired color from
TR17 0.2197 0.4141 0.4476 thermogram. In this figure, the masking, smoothing, and region fill-
TR18 0.2257 0.418 0.4717 ing process is used. The final segmented image of red color regions
TR19 0.2292 0.4203 0.4259 (hot regions) is shown in Fig. 10.
TR20 0.2305 0.4211 0.4163
This simulation is done over 500 sets of IR images. The geomet-
rical and statistical features of red color regions of ten sets of
images (TR10–TR20), out of 500 sets, are summarized in Tables 2
and 3. The data corresponding to set TR13 shows the abnormal val-
Ai ¼ Aib ð11Þ
ues of features from other sets, therefore, the system recognizes
these kinds of values, and generate alerts, if, these values crosses
where b is the blob number, n is total number of blobs in the seg- various set threshold values as given by ‘Proposed algorithm for
mented red color region, and Aib is the bth blob area at ith instant. temperature analysis in real time’.
In thermograms, a red color region indicates hotness of the
region. This implies that the rate of change of red color area
(DAR) is directly proportional to the rate of change of temperature Discussions
(DT) of that region of the object of desire as given in (12).
Rate of change of red color region area DAR ¼ dðAi A
i1 Þ
and This paper is focused to develop the non invasive monitoring
Rate of change of temperature DT ¼ dðT i T
i1 Þ and controlling systems for to detect early faults in electrical
Thus, assets.
The NIOLVIS is developed to deal with the off line health mon-
DAR / DT itoring and analysis of electrical systems such as: electrical con-
tacts, inductive heating, electrical utility substations, feeder
dðAi  Ai1 Þ dðT i  T i1 Þ poles, capacitor banks, electrical distribution panels, uninterrup-
) /
dt dt table power supplies, generator controls and transfer switches,
fuse cabinets, motor control centers, bus bars and ducts.
dðAi  Ai1 Þ dðT i  T i1 Þ
) ¼K ð12Þ The NIRTVMS is developed for continuous temperature moni-
dt dt toring of power transformer using infrared thermograms in real
where K is the constant of proportionality and it is termed as coef- time. This system can be used for the monitoring of thermal effects
ficient of infrared redness. The unit of K is pixel per Kelvin. Its value of large electrical machines and equipments. Also, this system is
increases with increase in emissivity and density of the material good for monitoring of the electrical assets where rise in tempera-

Table 4
Comparison of NIOLVIS and NIRTVMS with the existing system.

S. no. Parameters Existing systems NIOLVIS NIRTVMS

1 Type of monitoring Invasive Non invasive Non invasive
2 Speed of operation Slow Fast Very fast
3 Accuracy 90–95% 99.98% 98.46%
4 Human factor 30% 5% Nil
5 Environment factor 10% 1% 1%
6 Maintenance Frequently required Rarly required Rarly required
7 Running cost More Less Less
8 Flexibility of use in variety of applications Less More More
9 Direct effect on systems during fault Maximum Nil Nil
10 Reliability 80% 99% 98%
Z.A. Jaffery, A.K. Dubey / Electrical Power and Energy Systems 63 (2014) 753–759 759

ture is a prime constraint and direct measurement of temperature [12] Bortoni EC, Siniscalchi RT, Jardini JA. Determination of hydro generator
efficiency using infrared thermal imaging techniques. IEEE Trans. Energy
is not easily possible due to hazards conditions.
Convers 2011;26(4):1134–9.
A new temperature estimation algorithm is developed to esti- [13] Jaljal N, Trigeol JF, Lagonotte P. Reduced thermal model of an induction
mate the rate of temperature rise in the system with the rate of machine for real-time thermal monitoring. IEEE Trans Industr Electron
change of red color area in the thermogram. 2008;55(10):3535–42.
[14] Stipetic S, Kovacic M, Hanic Z, Vrazic M. Measurement of excitation winding
A comparison of the performance of the proposed systems are temperature on synchronous generator in rotation using infrared
given in Table 4 with that of existing invasive type monitoring sys- thermography. IEEE Trans Industr Electron 2012;59(5):2288–98.
tems. The parameters taken in this table are: type of monitoring [15] Meriaudeau F. Real time multispectral high temperature measurement:
application to control in the industry. Image Vis Comput 2007;25(6):1124–33.
system (invasive or non-invasive) [26,40,41]; speed of operation; [16] Newman TS, Jain AK. A survey of automated visual inspection. Comput Vis
accuracy [39]; human factor such as fatigue, vision level, and over- Image Underst 1995;61(2):231–62.
look of parameters; environmental factor [33] such as humidity, [17] Rosenfeld A. Image analysis and computer vision: 1999. Comput Vis Image
Underst 2000;78(2):222–302.
dust, extreme coldness or hotness, and rain; maintenance; running [18] Thacker NA, Clark AF, Barron JL, Beveridge JR, Courtney P, Crum WR, et al.
cost which includes repairing of the system, replacement of faulty Performance characterization in computer vision: a guide to best practices.
and damaged part due to incessant operation; flexibility of use in Comput Vis Image Underst 2008;109(3):305–84.
[19] Jaffery ZA, Dubey AK. Real time visual inspection system (RTVIS) for
variety of applications; direct effect on systems during fault, and calibration of mechanical gauges. In: IEEE international conference on recent
reliability [33,42–44]. The speed of operation and accuracy of the advances in intelligent computational system; 2011. p. 841–6.
systems depends upon the processor speed, type of cabling system, [20] Jaffery ZA, Dubey AK. Architecture of non-invasive real time visual monitoring
system for dial type measuring instrument. IEEE Sens J 2013;13(4):1236–44.
distance of camera from the host terminal, type of protocol, and
[21] Zhang H, Fritts JE, Goldman SA. Image segmentation evaluation: a survey of
medium of data transfer. unsupervised methods. Comput Vis Image Underst 2008;110(2):260–80.
[22] Crevier D. Image segmentation algorithm development using ground truth
image data sets. Comput Vis Image Underst 2008;112(2):143–59.
Conclusion [23] Weinland D, Ronfard R, Boyer E. A survey of vision-based methods for action
representation, segmentation and recognition. Comput Vis Image Underst
This research work has been conducted with a view to develop a [24] Guinness KM, O’Connor NE. Toward automated evaluation of interactive
noninvasive type system that can effectively monitor the temper- segmentation. Comput Vis Image Underst 2011;115(6):868–84.
ature of costly electrical assets and prevent the asset from catastro- [25] Fouquier G, Atif J, Bloch I. Sequential model-based segmentation and
recognition of image structures driven by visual features and spatial
phe through early fault detection. It is obvious that early relations. Comput Vis Image Underst 2012;116(1):146–65.
prevention is needed to prevent the faults in the electrical systems [26] Wu Y, Wang Y, Jia Y. Adaptive diffusion flow active contours for image
due to the thermal impact. Moreover, it may also improve the reli- segmentation. Comput Vis Image Underst 2013;117(10):1421–35.
[27] Otsu N. A threshold selection method from gray level histograms. IEEE Trans
ability of electrical power equipments. Syst, Man Cybern Electron, SMC 1979;9(1):62–6.
In this work, two non invasive systems: NIOLVIS and NIRTVMS [28] Gonzalez RC, Woods RE. Digital image processing. 3rd ed. New Delhi,
using thermogram images have been developed for the monitoring India: Pearson Education; 2011.
[29] Heidary K, Caulfield HJ. Pre smoothing effects in artificial color image
and controlling of off-line and real-time electrical power equip- segmentation. Comput Vis Image Underst 2013;117(3):195–201.
ments. Threshold based segmentation is used to detect the high [30] Vandenbroucke N, Macaire L, Postaire JG. Color image segmentation by pixel
thermal impact regions. The proposed system is simulated for a classification in an adapted hybrid color space. Application to soccer image
analysis. Comput Vis Image Underst 2003;90(2):190–216.
large number of thermogram images. The performance of the sys-
[31] Celik T, Tjahjadi T. Unsupervised color image segmentation using dual-tree
tems are analyzed and compared with the existing invasive type complex wavelet transforms. Comput Vis Image Underst 2010;114(7):813–26.
system. The results obtained are very promising and satisfactory. [32] Sengur A, Guo Y. Color texture image segmentation based on neutrosophic set
and wavelet transformation. Comput Vis Image Underst
References [33] Jadin MS, Taib S. Recent progress in diagnosing the reliability of electrical
equipment by using infrared thermography. Infrared Phys Technol
[1] Schneider J, Gaul AJ, Neumann C, Hogräfer J, Wellßow W, Schwan M, et al.
[34] Bagavathiappan S, Lahiri BB, Saravanan T, Philip J, Jayakumar T. Infrared
Asset management techniques. Int J Electr Power Energy Syst
thermography for condition monitoring – a review. Infrared Phys Technol
[2] Velasquez-Contrerasa JL, Sanz-Bobib MA, Arellanoa SG. General asset
[35] Teelen K, Veelaert P. Computing regions of interest for geometric features in
management model in the context of an electric utility: application to power
digital images. Discrete Appl Math 2009;157(16):3457–72.
transformers. Electr Power Syst Res 2011;81(11):2015–37.
[36] Kong SG, Heo J, Abidi BR, Paik J, Abidi MA. Recent advances in visual and
[3] McBee KD, Simões MG. Utilizing a smart grid monitoring system to improve
infrared face recognition—a review. Comput Vis Image Underst
voltage quality of customers. IEEE Trans Smart Grid 2012;3(2):738–43.
[4] Souahlia S, Bacha K, Chaari A. MLP neural network-based decision for power
[37] Falomir Z, Museros L, Abril LG, Escrig MT, Ortega JA. A model for the qualitative
transformers fault diagnosis using an improved combination of Rogers and
description of images based on visual and spatial features. Comput Vis Image
Doernenburg ratios DGA. Int J Electr Power Energy Syst 2012;43(1):1346–53.
Underst 2012;116(6):698–714.
[5] Chatterjee A, Bhattacharjee P, Roy NK, Kumbhakar P. Usage of nanotechnology
[38] Salem TE, Ibitayo D, Geil BR. Validation of infrared camera thermal
based gas sensor for health assessment and maintenance of transformers by
measurements on high-voltage power electronic components. IEEE Trans
DGA method. Int J Electr Power Energy Syst 2013;45(1):137–41.
Instrum Meas 2007;56(5):1973–8.
[6] Chatterjee A, Bhattacharjee P, Roy NK. Mathematical model for predicting the
[39] Chou YC, Yao L. Automatic diagnosis system of electrical equipment using
state of health of transformers and service methodology for enhancing their
infrared thermography. In: International conference on soft computing and
life. Int J Electr Power Energy Syst 2012;43(1):1487–94.
pattern recognition; 2009. p. 155–60.
[7] Silva ACM, Castro ARG, Miranda V. Transformer failure diagnosis by means of
[40] Asiegbu GO, Haidar AMA, Hawari K. Thermal defect analysis on transformer
fuzzy rules extracted from Kohonen self-organizing map. Int J Electr Power
using a RLC network and thermography. Circ Syst 2013;4:52–60.
Energy Syst 2012;43(1):1034–42.
[41] Zidat F, Lecointe JP, Morganti F, Brudny JF, Jacq T, Streiff F. Non invasive
[8] Dharanya S, Priyanka M, Rubini R, Umamakeswari A. Real time monitoring and
sensors for monitoring the efficiency of AC electrical rotating machines.
controlling of transformer. J Artif Intel 2013;6(1):33–42.
Sensors 2010;10(8):7874–95.
[9] Huda ASN, Taib SB, Ishak DB. Necessity of quantitative based thermographic
[42] Lizak F, Kolcun M. Improving reliability and decreasing losses of electrical
inspection of electrical equipments. PIERS 2012:1852–5.
system with infrared thermography. Acta Electrotechnica Et Informatica
[10] Huda ASN, Taib SB, Ishak DB. Analysis and prediction of temperature of
electrical equipment for infrared diagnosis considering emissivity and object
[43] Ciulavu C, Helerea E. Power transformer incipient faults monitoring. Ann Univ
to camera distance setting effect. PIERS 2012:1856–9.
Craiova-Electr Eng Ser 2008;32:72–7.
[11] Petrosyants I. IR thermography system for control and monitoring of energy
[44] Newport R. Improving motor reliability with infrared condition monitoring.
saving and safety in heating and electrical equipment of housing services. In:
Electr Line 2002;8(1):36–44.
International conference on quantitative infrared thermography; 2012. p. 1–9.