Fault-Level Grading of Photovoltaic Cells Employing Lightweight Deep Learning Models.

Area of Science:

• Renewable Energy Systems
• Artificial Intelligence in Engineering
• Materials Science

Background:

• Photovoltaic (PV) cell deployment is increasing, necessitating efficient fault detection.
• Accurate fault-level grading is crucial for timely PV cell repair and replacement.
• Existing methods may lack the speed and accuracy required for real-time PV systems.

Purpose of the Study:

• To develop and evaluate deep neural network models for fault-level grading of PV cells.
• To create lightweight, computationally efficient models for real-time fault detection.
• To compare the performance of different deep learning architectures on PV cell defect classification.

Main Methods:

• Utilized a dataset of 2,624 electroluminescence images of PV cells with four defect levels.
• Developed classical artificial neural networks using hand-crafted texture features.
• Designed optimized, lightweight convolutional neural network (CNN) architectures for real-time processing.
• Conducted binary and multiclass classification experiments.

Main Results:

• The proposed CNN model achieved higher accuracy than state-of-the-art in binary classification with reduced computational complexity.
• A CPU-based model demonstrated superior accuracy and a significantly lighter architecture compared to GPU-based solutions.
• State-of-the-art results were obtained in multiclass categorization with 83.5% accuracy.
• The models provide efficient, computationally inexpensive, CPU-based solutions.

Conclusions:

• Deep neural networks, particularly lightweight CNNs, are effective for real-time PV cell fault-level grading.
• The developed models offer a practical and efficient solution for maintaining the reliability of PV energy systems.
• This research contributes to advancing automated diagnostics in the renewable energy sector.