Identifying defective solar cells in electroluminescence images using deep feature representations

Area of Science:

• Materials Science
• Electrical Engineering
• Computer Science

Background:

• Electroluminescence (EL) imaging is crucial for identifying surface defects in photovoltaic (PV) modules.
• Manual analysis of EL images is subjective, time-consuming, and requires specialized expertise.
• Automated defect detection in PV modules is essential for efficient quality control.

Purpose of the Study:

• To develop a hybrid, fully-automated classification system for detecting diverse defects in EL images.
• To enhance the accuracy and efficiency of defect analysis in PV modules.
• To overcome the limitations of manual inspection in EL image analysis.

Main Methods:

• A hybrid deep learning approach fusing Inception-V3 and ResNet50 models was employed.
• Deep feature representations were extracted and combined for discriminative feature vectors.
• A classifier layer was utilized for defect categorization.

Main Results:

• The system achieved high accuracy in both binary (98.15%) and multi-class (95.35%) classification tasks.
• Defect detection was performed rapidly, with processing times under 1 second per image.
• The system demonstrated robust performance on a large dataset of 2,624 EL images.

Conclusions:

• The proposed automated system significantly improves the accuracy and speed of defect detection in EL images.
• This AI-driven approach offers a reliable and efficient alternative to manual inspection for PV module quality assurance.
• The developed system has the potential to streamline manufacturing processes and enhance the reliability of solar energy technologies.