Volume 8, Issue 5, April – 2023 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Fault Identification in Solar PV Panels Using Thermal

Image Processing Technique
Dr. Renuka Devi S M1 S.Keerthana2
1 2
Professor, Dept. of ECE, G. Narayanamma Institute of Dept. of ECE G. Narayanamma Institute of Technology and
Technology and Science (for women) Hyderabad, India Science (for women) Hyderabad, India

B.Akhila3 G. Meghana4
3 4
Dept. of ECE G. Narayanamma Institute of Technology and Dept. of ECE G. Narayanamma Institute of Technology and
Science (for women) Hyderabad, India Science (for women) Hyderabad, India

K. Meghana5
5
Dept. of ECE G. Narayanamma Institute of Technology and Science (for women) Hyderabad, India

Abstract:- Photovoltaic (PV) module monitoringand brought on by lower PRs. The inspection methods focus on
upkeep are essential for a dependable and effective isolating performancedeteriorations rather than assessing the
operation. Due to hotspots in PV modules brought on by dependability of PV modules, which requires an
a variety of flaws and operational issues, the understanding of the different sorts of faults that can occur
dependability of the PV system may be put in jeopardy. when the system is in use. Large PV facilities cannot be
Hotspots should be found and categorized from a monitored using visual inspections and human
monitoring perspective for later maintenance. In this interpretation. Many contemporary systems incorporate
study, hotspots are identified, assessed, and categorized image processing to automate the procedure whereby
using thermal pictures of PV modules and a machine broken PV modules may be promptly found or categorized
learning technique. To do this, categorization is based on into the appropriate categories.
the texture and histogram of gradient (HOG) features of
thermal pictures of PV modules. The machine learning II. LITERATURE SURVEY
method like Naive Bayes (nBayes) classifier is used to
train the images in order to identify the hotspots and Different approaches to fault detection and analysis,
classifies them into defective and non-defective images. including monitoring systems, artificial intelligence based I
and C analysis, voltage and current measurements, and
Keywords:- Hotspots, Monitoring, Photovoltaic (PV) power loss measurements, lack precise data and are not
Modules, Naive Bayes Classifier, Texture and Histogram of decentralized in the system for inverters and fault diagnosis
Gradients (HOG) Thermal Images. [1]. Assessments of the fully operational power plants were
made after evaluating the solar PV system's connections to
I. INTRODUCTION the grid and its simulation using the MATLAB software [2].
Comparisons between the regular model and anticipated
Due in large part to the continuing decline in PV values of the proposed system were made using the output
module prices and the growing interest in environmentally readings for the fault system on PV from the various types of
friendly energy sources, solar PV energy generation and shading. A graph showing the noise for these outputs'
utilization have increased significantly over the past few detection times and a stability curve for the relevant data
decades. The use of solar PV modules is widespread, with were also made [3].
examples including utility-scale power plants, remote stand-
alone telecommunication stations,residential rooftops, and The spectral information is supplemented by thetexture
nearly zero energy building facades. The International properties. HOG features are utilized for picture identification
Energy Agency (IEA) estimates that there are currently and image detection by storing the local gradients in the
more than 400 GWP of installed solar powersystems images [4]. Histogram equalization is carried out to increase
globally, with crystalline silicon (c-Si), both mono and poly, the contrast to facilitate picture distinction [5]. Modelling a
accounting for more than90%. genuine PV system is extremely difficult since electrical
characteristics differ greatly between PV systems due to
According to several studies that show lower variations in the location, physical layout, and PV module
performance ratios (PR) under actual conditions, the energy manufacture (including size, material, and ground
production of a PV system is typically significantly lower connection) [6].
than the anticipated performance. Degradation in PV
module performance that often begins after the first few
years of operation and getting worse over time is frequently

IJISRT23APR2277 www.ijisrt.com 2400

Volume 8, Issue 5, April – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Numerous elements, including wear-and-tear issues, III. PROPOSED SYSTEM
tracking errors for the maximum power point, electrical wire
losses, production issues, and overheating, can have an The proposed system consists of Thermal Camera
impact on the energy losses and output power failures in the which capture the image and process the image using
PV system. A PV system's output power may be reduced by MATLAB software and detects the fault in the solar panel.
these various factors by 50% [7]. Modern technologies and To efficiently monitor and classify solar PV modules, a
nondestructive testing techniques like the thermal image machine learning- based technique is proposed. Essential
process are used to identify faults in solar PV modules. To features are extracted from the non-radiometric thermal
achieve perfection for the deduction of the fault, a neural images of PV modules and redundant information is
network classifier-based method is designed using various removed, which forms the basis of the classification. The
sets of criteria and collections of modules. [8]. algorithm is trained using the multi-class density-based
classifier, i.e., a Naive Bayes (nBayes) classifier, instead of
Current and voltage are the two factors that have the a binary classifier. Notably, the nBayes classifier can solve
biggest impact on a PV system's performance. The electrical the multi-class density-based classification problems, i.e., to
signature of each defective module and array was repaired calculate explicit probabilities for each hypothesis based on
using a straightforward current-voltage analysis method that the Bayes theorem. In nBayes classifiers, the value of a
took into account the deformations generated on the I/V specific feature is autonomous to the value of any other
curves [9]. For the industrial control computer to identify feature, given to the class variable. According to the training
each solar module individually, they must be treated as method, PV modules are categorized into three types:
single cells. Then, for the various types of issues, including a
broken grid, fragmented cells, black fragments on solar cells,  Defective
and cracks in the cells, several approaches are utilized to  Non-Defective
pinpoint theissue and determine the degree of faults present.
It offers extremely precise and quick real-time processing
[10]

IV. WORKING

 The Block Diagram for the Proposed System:

Fig 1 The Proposed System

IJISRT23APR2277 www.ijisrt.com 2401

Volume 8, Issue 5, April – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
 Modules:  Training:
Thermal pictures are used as training data for these
 Data Acquisition three distinct types (Defective, NDH, and NDNH). In order
 Pre-Processing to train the model, 4 the HOG and texture features are
 Feature Extraction computed, resulting in a concatenated feature descriptor
 Training foreach training image.
 Classification
 Classifier:
 Module Explanation: An n-Bayes-based multi-class density-based classifier
from the library of pattern recognition tools is used to
categorize solar PV modules into the groups of faulty,
 Data Acquisition: NDH,and NDNH. This classifier has several features, and
The FLIR thermal camera has captured non-enhanced each training PV thermal picture affects the probability of a
and nonradiometric thermal photos of solar PV modules. given hypothesis. This allows the classifier to more
Theacquired dataset was manually labelled and divided into effectively rule out any incompatible hypotheses.
the three classifications of Defective, NDH, and NDNH by Combining these forecasts based on their confidence score
the PV module. also assigns a confidence percentage to the projections.
Additionally, it explicitly resolves the problem of
 Pre-Processing: classifying PV modules into their appropriate groups by
Pre-processing is a necessary step to improve the classifying each fresh thermal image of a PV module as a
textural information and dependability of the data gathered prediction function of several hypotheses that are weighted
from thermal pictures of PV modules. The suggested n- by their probabilities.
Bayes- based classification approach uses the texture and
HOG characteristics. The RGB image is preprocessed by V. RESULTS AND DISCUSSION
making it a grayscale image to measure the amount of light
that is present at each pixel in a single band of the The thermal images dataset was created using a solar
electromagnetic spectrum. Additionally, to enhance the panel of 12 volts rating and Seek Thermal Camera (CT-
grayscale PV's contrast. AAA). The small area of the solar panel was made inactive
by covering it with a cardboard. This is conceptually the
 Feature Extraction: effect similar to that of hotspot. The 200 images were taken,
When the categorization is based on machine learning with 100images each of normal PV panels and hotspot
techniques, features are crucial. Then, from each thermal images. 170images are used for training and 30 images for
image of a PV module, the texture and HOG features are testing. The sample images are shown infigure 2.
retrieved. For the purpose of identifying and categorizing
spectrally diverse images, such as thermal images of PV
modules, the texture feature is utilized to augment the
spectral information. By storing the local gradients in the
images, the HOG feature, on the other hand, is used for
image identification and image detection. In this research,
the feature descriptors for the categorization of solar PV
modules are built using the two categories of characteristics.
Which are:

 Texture Features
 HOG Attributes Fig 2 Thermal Images of Defective and Non-Defective
Solar Panels

In the following equation, i and j stand for the

positions of pixels inside an image, mij is the value of those
pixels, N stands for the number of grey levels in the image, Fig 3 Pre-processing of a defective solar panelwhich
is the mean, and 2 stands for the variance of the pixel includes, Gray scale image, HOG feature extraction and
values. Histogram Equalization of an image.

IJISRT23APR2277 www.ijisrt.com 2402

Volume 8, Issue 5, April – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
 The feature extraction of Contrast, Correlation, Energy REFERENCES
and Homogeneity maps for Defective andNon-Defective
solar panel are given in Figure 4 and Figure 5. [1]. Hindawi: International Journal of Photoenergy
Volume 2022, Infrared thermal images of solar panel
for fault identification using thermal image
processing technique’, Article ID 6427076.
[2]. v. Vi, k. Raja, v. S. Chandra sekar, and t. Ramkumar,
“thrust force evaluation and microstructure
characterization of hybrid composites
(al7075/b4c/bn) processed by conventional casting
technique,” journal of the brazilian society of
mechanical sciences and engineering, vol. 41, no. 5,
p. 228, 2019.
[3]. E. M. Natsheh and A. Albarbar, “Solar power plant
performance evaluation: simulation and experimental
validation,” Journal of Physics: Conference Series,
vol. 364, article 012122,2012.
[4]. Renuka Devi S.M, Anusha Akula, “Automatic Speed-
Limit Sign Detection and Recognition for Advanced
Fig 4 Feature extraction of Defective solar panel Driver Assistance Systems”, Volume-8 ISSN: 2278-
3075,2019.
[5]. Kumar R, V. Chandrasekhar, and Anil Kumar Rao,
“An automatic multi-threshold image processing
technique for mushroom disease segmentation",
Volume-7, ISSN: 2393-83742020.
[6]. Barun Basnet, Hyun Jun Chun, Junho Bang, "An
Intelligent Fault Detection Model for Fault Detection
in Photovoltaic Systems", Journal of Sensors, vol.
2020, Article ID 6960328, 11 pages, 2020.
[7]. Muhammad Umair Ali, Hafiz Farhaj Khan, Manzar
Masud, Karam Dad Kallu, Amad Zafar, A machine
learning framework to identify the hotspot in
photovoltaic module using infrared thermography,
Solar Energy, Volume 208,2020, ISSN 0038-092X.
[8]. Y. Yang, Q. Chen, and Y. Wan, “A fast near-
optimum block truncation coding method using a
truncated K -means algorithm and inter-block
correlation,” AEU-International Journal of
Fig 5 Feature extraction of Non-Defective.solar panel Electronics and Communications, vol. 65, no. 6, pp.
576–581, 2011.
[9]. M. H. Ali, A. Rabhi, A. El Hajjaji, and G. M. Tina,
“Real time fault detection in photovoltaic systems,”
Energy Procedia, vol. 111, pp. 914–923, 2017.
[10]. S. Guesmi, K. Jamoussi, and M. Ghariani,
“Efficiency improvement of PV module using an
electrical thermal model,” Journal of Electrical
Systems, vol. 16, no. 2, pp. 246–256, 2020.

Fig 6 Confusion Matrix of Given Dataset

IJISRT23APR2277 www.ijisrt.com 2403