Open AccessFault Detection in aerial images of photovoltaic modules based on Deep learning
Author(s) -
S. Venkatesh,
V. Sugumaran
Publication year - 2021
Publication title -
iop conference series. materials science and engineering
Language(s) - English
Resource type - Journals
eISSN - 1757-899X
pISSN - 1757-8981
DOI - 10.1088/1757-899x/1012/1/012030
Subject(s) - photovoltaic system , fault detection and isolation , softmax function , convolutional neural network , computer science , deep learning , artificial intelligence , fault (geology) , aerial image , feature extraction , test bench , real time computing , feature (linguistics) , pattern recognition (psychology) , computer vision , engineering , embedded system , image (mathematics) , electrical engineering , geology , seismology , actuator , linguistics , philosophy
Operation and maintenance of photovoltaic (PV) modules are currently the prime concerns of the expanding photovoltaic industry. Unmanned aerial vehicles (UAVs) are applied in the field of inspection and monitoring of faults that occur in a photovoltaic module (PVM). Such inspections can significantly reduce time and human interference to provide accurate classification results. Technological advancements and innovative techniques in the fast moving world expect instantaneous results. Fault diagnosis is one such technique that provides instantaneous results and assures enhanced lifetime of various critical components. This paper presents the fault detection in PVM based on deep learning with the help of aerial images acquired from UAVs. Convolutional neural networks (CNN) are adopted to extract high level features from the images which are classified using the softmax activation function. The feature extraction and fault classification is carried out by using a pre-trained VGG16 network. A total of six test conditions are considered in the study. Burn marks, delamination, discoloration, glass breakage, good panel and snail trail are the several test conditions considered. The classification result for the pre-trained CNN model is exhibited and performance of the model is evaluated.