No abstract

No abstract

During the production of energy using photovoltaic (PV) panels, solar cells may be affected by different environmental aspects, which cause many defects in the solar cells. Such defects should be identified and categorized to improve energy productivity. Most recent studies developed many deep learning models to categorize solar cell defects, whereas tiny defects were not categorized due to the blurry and overlapping edges. Hence, this paper develops an effective deep-learning model of cell partitioning and defect classification to satisfy the criteria for solar panel manufacturing monitoring. At first, the low-resolution solar cell images are enriched into super-resolved images using the improved super-resolved adversarial network. Then, those images are passed to the novel deep feature discrimination network, which extracts fault-related characteristics from the solar cell images. Those characteristics are further fed to the softmax classifier to categorize the defected or non-defected solar cell images. Moreover, a new mixture of error functions is used for fine-tuning and rapid convergence. Further, a detection bias and compensation in missed faults that limit the model accuracy are found to be unaffected by the categorization of defected/non-defected panels in a real-time manufacturing unit. Finally, assigning certain forecasting standards for various faults categories, the overall test analysis using solar cell images on 5 days exhibits that the presented model attains a maximum accuracy of 99.91% compared to the classical deep learning models. So, this model has a high probability to be used in a real manufacturing unit.