Solar modules are subject to a range of atmospheric events such as rain, wind, and snow and for this reason, they are usually built with protection frames. Nevertheless, these measures are insufficient to prevent damages, especially from the mechanical ones (e.g., the fall of tree branches) decreasing the power efficiency of solar modules. Then it is necessary to monitor their healthy conditions and replace or repair defective units. Electroluminescence, a useful inspection modality of solar modules, makes it possible to detect even the finest defects on the surface of solar modules. However, the analysis of these images is usually carried out by human operators, making this inspection practice expensive, time-consuming and it requires very specific knowledge. In state-of-the-art there are several works that distinguish between a healthy cell and defective cell, but a public dataset of possible defects in solar cells has never been published. For this reason, we propose a new dataset and a preliminary benchmark to make an automatic and accurate classification of defects in solar cells. The dataset includes five classes of defects and the pre-trained ResNext50 network reaches 0.07 Hamming Distance.

E-ELPV: Extended ELPV Dataset for Accurate Solar Cells Defect Classification

Grisanti M.;Ortis A.;Battiato S.
2024-01-01

Abstract

Solar modules are subject to a range of atmospheric events such as rain, wind, and snow and for this reason, they are usually built with protection frames. Nevertheless, these measures are insufficient to prevent damages, especially from the mechanical ones (e.g., the fall of tree branches) decreasing the power efficiency of solar modules. Then it is necessary to monitor their healthy conditions and replace or repair defective units. Electroluminescence, a useful inspection modality of solar modules, makes it possible to detect even the finest defects on the surface of solar modules. However, the analysis of these images is usually carried out by human operators, making this inspection practice expensive, time-consuming and it requires very specific knowledge. In state-of-the-art there are several works that distinguish between a healthy cell and defective cell, but a public dataset of possible defects in solar cells has never been published. For this reason, we propose a new dataset and a preliminary benchmark to make an automatic and accurate classification of defects in solar cells. The dataset includes five classes of defects and the pre-trained ResNext50 network reaches 0.07 Hamming Distance.
2024
978-3-031-47720-1
978-3-031-47721-8
Convolutional neural networks
Deep learning
Defect classification
Defect detection
Electroluminescence imaging
Solar energy
Solar modules
Visual inspection
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/20.500.11769/588149
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 0
  • ???jsp.display-item.citation.isi??? ND
social impact