Fault detection and computation of power in PV cells under faulty conditions using deep-learning

Renewable energy is considered to be an alternate option for limiting the consumption of fossil fuel along with reducing environmental pollution. Among the possible renewable energy resources, solar energy is considered to be the key candidate as it is the most economical and energy efficient. Keeping in view its importance, there has been an exponential increase in energy harvesting using photovoltaic (PV) systems across the globe in recent years. However, to ensure optimal performance, system health monitoring is essential for such a system. Current traditional monitoring methods are laborious, expensive, time-consuming, and prone to error. To address the limitations of these approaches, the proposed work follows two main approaches. First, an effective deep-learning method is proposed for the identification of the types of cracks in the PV cell such as microcracks and deep cracks. In microcracks, the crack's orientation is crucial and therefore classified accordingly. Next, the power analysis is performed based on the severity of the cracks. In case of deep cracks, it is observed that the output power efficiency is proportional to crack size. For crack identification, four deep learning models, namely U-net, LinkNet, FPN, and attention U-net, are trained, evaluated, and compared using an online public dataset of electroluminescence images. The efficiency of the models is evaluated using different metrics, including intersection over union (IoU) and F1-Score. These models are then subjected to an ensemble learning technique, which results in an accurate and robust segmentation, IoU, and F1-Score.