Vision Transformer-Based Fault Diagnosis for Solar Panels Using Visible Image Dataset

Solar photovoltaic (PV) systems have become one of the most well-known means of renewable energy in recent years. However, the faults in solar panels, such as cracks, delamination, discoloration, shading, etc., significantly reduce the plant's efficiency and lifetime, ultimately resulting in expensive repair bills if they go undetected. Traditional fault detection techniques, such as manual inspections and electrical testing, were known to be expensive, time-consuming, and prone to human error.In this paper, a deep-learning novel architecture, vision transformer, has been introduced for the detection and diagnosis of the four most critical faults in solar PV panels.The experimental results show that the vision transformer model trained on the solar image dataset achieved a remarkable accuracy of 93.73% with an F1 score of 93.72%. The performance of the proposed model is evaluated with other state-of-the-art deep learning (DL) models, such as EfficientNetV2, MobileNetV3, and ResNet50v2, and obtained accuracy of 72.75%, 92.04%, and 91.55%, respectively. The outcome shows that the suggested approach has a great deal of potential for diagnosing solar PV system errors, which would help sustainably and efficiently run renewable energy facilities.