Advanced Deep Learning Solutions for Automated Diagnosis of Solar Panel Issues
Chaitra Pala, Poojitha Bollem, Nizampatnam Neelima · 2024
Solar energy stands out as a highly reliable renewable technology with global applicability. The growing integration of solar photovoltaic systems worldwide emphasizes the need for sustaining peak performance and maximizing energy output. Despite their widespread feasibility, solar panels face various vulnerabilities, including dust accumulation, snow, bird droppings, and other anomalies that can compromise module efficiency and overall energy production. Because of this cleaning solar panels has become a critical task. Leveraging the effectiveness of Convolutional Neural Networks (CNNs) in image classification, particularly for visual inspection tasks, this paper compares the results of CNN VGG16 and RESNET50 architectures in detecting faults in solar panels using thermal images. Kaggle's image dataset encompasses various fault types such as dust, snow, bird droppings, and physical and electrical damage on solar panel surfaces. The suggested VGG16 CNN model attains a noteworthy success rate of 95.4%, while RESNET50 achieves a comparatively lower accuracy of 83%. Notably, this paper distinguishes itself by expanding the scope of previous works, which typically focused on detecting three fault classes. In contrast, the proposed model successfully identifies six fault classes, using both architectures significantly advancing solar panel fault detection.