Automatic detection of photovoltaic module defects in infrared images with isolated and develop-model transfer deep learning

[Display omitted] •State-of-the-art framework is proposed for automatic defect detection in PV modules.•Infrared images dataset of normal operating and defective PV modules is collected.•Isolated and develop-model transfer deep learning frameworks are proposed.•Isolated & transfer learned method...

Full description

Saved in:
Bibliographic Details
Published in:Solar energy Vol. 198; pp. 175 - 186
Main Authors: Akram, M. Waqar, Li, Guiqiang, Jin, Yi, Chen, Xiao, Zhu, Changan, Ahmad, Ashfaq
Format: Journal Article
Language:English
Published: New York Elsevier Ltd 01.03.2020
Pergamon Press Inc
Subjects:
Accuracy
Artificial neural networks
Automatic defect detection
Automation
Datasets
Deep learning
Defects
Develop-model transfer deep learning
Electroluminescence
Infrared imagery
Infrared images
Infrared imaging
Inspection
Isolated deep learning
Labeling
Machine learning
Modules
Monitoring methods
Neural networks
Performance enhancement
Photovoltaic (PV) modules
Photovoltaic cells
Photovoltaics
Solar energy
Thermography
Transfer learning
Isolated deep learning
Automatic defect detection
Develop-model transfer deep learning
Photovoltaic (PV) modules
Infrared images
Thermography
ISSN:0038-092X, 1471-1257
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:[Display omitted] •State-of-the-art framework is proposed for automatic defect detection in PV modules.•Infrared images dataset of normal operating and defective PV modules is collected.•Isolated and develop-model transfer deep learning frameworks are proposed.•Isolated & transfer learned methods give 98.67% and 99.23% accuracy respectively.•These frameworks are qualitatively evaluated with experimental testing. With the rising use of photovoltaic and ongoing installation of large-scale photovoltaic systems worldwide, the automation of photovoltaic monitoring methods becomes important, as manual/visual inspection has limited applications. This research work deals with automatic detection of photovoltaic module defects in Infrared images with isolated deep learning and develop-model transfer deep learning techniques. An Infrared images dataset containing infrared images of normal operating and defective modules is collected and used to train the networks. The dataset is obtained from Infrared imaging performed on normal operating and defective photovoltaic modules with lab induced defects. An isolated learned model is trained from scratch using a light convolutional neural network design that achieved an average accuracy of 98.67%. For transfer learning, a base model is first developed (pre-trained) from electroluminescence images dataset of photovoltaic cells and then fine-tuned on infrared images dataset, that achieved an average accuracy of 99.23%. Both frameworks require low computation power and less time; and can be implemented with ordinary hardware. They also maintained real time prediction speed. The comparison shows that the develop-model transfer learning technique can help to improve the performance. In addition, we reviewed different kind of defects detectable from infrared imaging of photovoltaic modules, that can help in manual labelling for identifying different defect categories upon access to new huge data in future studies. Last of all, the presented frameworks are applied for experimental testing and qualitative evaluation.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0038-092X
1471-1257
DOI:10.1016/j.solener.2020.01.055