Research Article
Classification Panel Defects from Infrared Thermal Images Using Deep Neural Networks
Mabikana Voula Boniface Herve*
,
Mabiala Louboto Antoine Victorien,
Nkombo Mazouka Michel,
M’Passi Mabiala Bernard
Issue:
Volume 12, Issue 4, August 2026
Pages:
59-69
Received:
23 June 2026
Accepted:
11 July 2026
Published:
11 August 2026
DOI:
10.11648/j.ajme.20261204.11
Downloads:
Views:
Abstract: Development research has today, through different technologies, boosted the solar panel manufacturing industry. These technologies allow these solar panels to make the most of solar radiation to produce electricity, and as a result, improve efficiency. These photovoltaic solar panels, whenever they are installed, often have their performance affected by faults and anomalies that lead to a drop in energy efficiency. Many studies have looked into diagnosing, detecting, and locating these faults using algorithms. Building on this research and the experience gained from these systems in operation, several faults have been identified in photovoltaic cells. The goal of this study was to design, based on these different defects, artificial intelligence algorithms to detect, classify, and predict defects on solar panels over time. The model was trained using a neural network with the public database available on GitHub (intended for researchers, public services, solar project developers, and decision-makers). It contains infrared imaging data highlighting anomalies in a photovoltaic solar field. This dataset, made up of 20,000 images, allowed us to create an approximator (neural network) linking the pixels of a thermal image of a photovoltaic module to the presence of defects. During model training, the decrease in training and validation losses shows that this model has learned and can improve its performance. Beyond 20 epochs, the model converges, which means that no additional learning can improve the model's performance. This developed, trained, and validated model was exported to Simulink and allowed us to classify and predict solar panel faults, with an average accuracy of over 95% for some faults and 100% for others.
Abstract: Development research has today, through different technologies, boosted the solar panel manufacturing industry. These technologies allow these solar panels to make the most of solar radiation to produce electricity, and as a result, improve efficiency. These photovoltaic solar panels, whenever they are installed, often have their performance affect...
Show More
