1. DETECTION OF FAULTY PHOTOVOLTAIC
PANELS USING MACHINE LEARNING
Supervisor
Prof. Dr. Aamir Hanif
Chairperson-EE
Co- Supervisor
Dr. Amad Zafar
Associate Professor -EE
Presented By
Waqas Ahmed
UW-18-EE-MS-001
3. Problem Statement
Faulty photovoltaic (PV) panels connected with healthy PV panels
compromise the overall system’s performance and efficiency.
Therefore, timely identification and classification of faulty PV panels
from healthy PV panels is necessary.
4. Objective
The objective of this research work is to identify and classify faulty PV
panel(s) to restrain the power losses.
5. PV System
PV systems are preferred choice. Therefore, PV system manufacturer has to
guarantee the long-term performance of PV panels for energy production, which
may suffer multiple issues such as overheating [Umair et al.,].
PV panel efficiency reduces over the
time, depending upon operational
conditions such as panel type,
environmental conditions etc. [BLD 290
Watt]
Age Output
10 years 90%
25 years 80%
6. PHOTOVOLTAIC FAULTS
Basnet, B., Chun, H., et al., Journal of Sensors, 2020.
Uneven shading
PV panels shading due to bird drop, dust, clouds, vicinal environment object
Hotspot
Area of high temperature that affect only one zone of the solar panel and result in a localized
decrease in efficiency
Cracks
Glass breakdown, protection breakdown
Aging
Environmental, operational and time collaborative effect
7. FAULTS IDENTIFICATION METHODS
Inspection of a large scale PV system is difficult and time consuming, which
needs upgradation/automation [Umair et al., Wuqin et al., Basnet et al.,]
For inspection, two main methods to identify the PV system faults i.e.,
1. Electrical signals (V/I characteristics) [Basnet et al.]
2. PV thermographic images (Electroluminescence [Wuqin et al.], infrared
[Umair et al.] etc.)
8. FAULTS IDENTIFICATION METHODS
V/I Characteristics Electro
-luminescence
Infrared
Thermography
Pros Real time data
monitoring
High resolution
images, better results,
non destructive
computational,
time, storage low
Cons Sensors, may
consider healthy PV
as faulty in winters,
costly
Night environment,
computational, time,
storage high
Low resolution
images, lower fault
detection results
9. GENERAL FRAMEWORK
Dunderdale, C., Brettenny, et al., Progress in Photovoltaics: Research and Applications, 2020.
Acquire data for
Analysis
Clean and Prepare
Data
Image Feature
Calculations
(if Applicable)
Apply Machine
Learning Algorithm
Cross Validate
Results
Conclusion
10. Literature Review
Reference Machine
Learning
Approach
Inspection
Method
Features Identification
Classes
Ali, M. U., Khan, H. F.,
Masud, M. et.al., 2020
Support Vector
Machine
Infrared Thermography
Homogeneity, energy,
contrast, correlation,
HOG, RGB LBP
3 Classes (faulty,
hotspot healthy and
healthy)
Akram, M. W., Li, G.,
Jin, Y., Chen. et.al.,
2020
Light Convolutional
Neural Network
Trained on
Electroluminescence,
transfer learning to
Infrared
Thermography
-
2 Classes
(Faulty and healthy)
11. Continue...
Reference Machine Learning
Approach
Inspection
Method
Features Identification
Classes
Tang, W., Yang, Q.,
Xiong, K.et.al., 2020
CNN Electroluminescence - Two
Gallardo-Saavedra,
S., Hernández-
Callejo, L., Alonso-
García et.al., 2020
-
Infrared Thermography
Electroluminescence
V/I Characterisitcs
Power inverter
with bidirectional
power flow
capability
Faults (mainly two)
Dunderdale, C.,
Brettenny, W.,
Clohessy, C. et.al.,
2019
Scale invariant feature transform
with random forest classifier
Infrared Thermography
Local feature
extraction,
codebook
generation, bag of
visual words
Five fault types
14. References
Ali, M. U., Khan, H. F., Masud, M., Kallu, K. D., & Zafar, A. (2020). A machine learning framework to identify the hotspot in photovoltaic module
using infrared thermography. Solar Energy, 208, 643-651.
BLD 290-Watt solar Panels: https://www.enfsolar.com/pv/panel-datasheet/crystalline/23907
Tang, W., Yang, Q., Xiong, K., & Yan, W. (2020). Deep learning based automatic defect identification of photovoltaic module using
electroluminescence images. Solar Energy, 201, 453-460.
Basnet, B., Chun, H., & Bang, J. (2020). An Intelligent Fault Detection Model for Fault Detection in Photovoltaic Systems. Journal of Sensors,
2020.
Akram, M. W., Li, G., Jin, Y., Chen, X., Zhu, C., & Ahmad, A. (2020). Automatic detection of photovoltaic module defects in infrared images with
isolated and develop-model transfer deep learning. Solar Energy, 198, 175-186.
Gallardo-Saavedra, S., Hernández-Callejo, L., Alonso-García, M. D. C., Muñoz-Cruzado-Alba, J., & Ballestín-Fuertes, J. (2020). Infrared
Thermography for the Detection and Characterization of Photovoltaic Defects: Comparison between Illumination and Dark Conditions. Sensors,
20(16), 4395.
Dunderdale, C., Brettenny, W., Clohessy, C., & van Dyk, E. E. (2020). Photovoltaic defect classification through thermal infrared imaging using a
machine learning approach. Progress in Photovoltaics: Research and Applications, 28(3), 177-188.
Ali, M. U., Khan, H. F., Masud, M., Kallu, K. D., & Zafar, A. (2020). A machine learning framework to identify the hotspot in photovoltaic module using infrared thermography. Solar Energy, 208, 643-651.
Approach: SVM
3 classification issue
Accuracy 92%
Features Texture, HOG, RBG and LBP
BLD 290 Watt solar Panels:
https://www.enfsolar.com/pv/panel-datasheet/crystalline/23907
Approach: Probabilistic Neural Network
Claimed efficiency: 100%
Features: V/I characteristics
Issues Condition: faulty performance and healthy performance in winter is same due to low irradiance
Method Machine learning
Ali, M. U., Khan, H. F., Masud, M., Kallu, K. D., & Zafar, A. (2020). A machine learning framework to identify the hotspot in photovoltaic module using infrared thermography. Solar Energy, 208, 643-651.
Approach: SVM
3 classification issue
Accuracy 92%
Features Texture, HOG, RBG and LBP
Tang, W., Yang, Q., Xiong, K., & Yan, W. (2020). Deep learning based automatic defect identification of photovoltaic module using electroluminescence images. Solar Energy, 201, 453-460.
CNN Apporach
Features by CNN itself
Accuracy 84% Defect free
Accuracy 82% Micro crack
Basnet, B., Chun, H., & Bang, J. (2020). An Intelligent Fault Detection Model for Fault Detection in Photovoltaic Systems. Journal of Sensors, 2020.
Approach: Probabilistic Neural Network
Claimed efficiency: 100%
Features: V/I characteristics
Issues Condition: faulty performance and healthy performance in winter is same due to low irradiance
Method Machine learning
M. Waqar self induced lab faults, real time issues missing like assumed microcracks of >3cm, below can’t be identified.