Key points are not available for this paper at this time.
Abstract. Electroluminescence (EL) imaging of photovoltaic solar cells can detect and classify solar panel faults. This method allows technicians and manufacturers to identify defective panels that may affect performance and longevity. However, noise in EL images and solar cell silicon granularity make this process difficult. The paper presents an automated deep-learning framework to identify faulty and normal solar cells from images. Xception, a popular CNN network, is modified to reduce complexity and solve overfitting issues. Few separable convolution layers were removed from the original Xception network, and lateral dropout layers were added. The proposed deep CNN is tested on ELVP. To balance two classes, images are augmented with two rotations and dimensional shifting. Finally, the proposed model is compared to a pretrained CNN network and leading methods. The quantitative analysis showed that the model performed better than previous methods, with 94.382% accuracy, 92% precision, 95.12% recall rate, and 93.53% F1 score. Module fault identification helps with maintenance planning. Solar energy's widespread adoption and growth as a renewable and sustainable power source may result.
Hiren Mewada (Tue,) studied this question.