Automated Micro-Crack Detection within Photovoltaic Manufacturing Facility via Ground Modelling for a Regularized...

Area of Science:

• Materials Science
• Electrical Engineering
• Computer Vision

Background:

• Photovoltaic cell manufacturing involves high temperatures and pressure, leading to surface defects like micro-cracks.
• Manual inspection of these defects is prone to human error, bias, and high costs.
• Automated defect detection is crucial for improving solar cell quality and production efficiency.

Purpose of the Study:

• To develop an automated, non-invasive method for detecting micro-cracks on photovoltaic cell surfaces.
• To model cell surfaces with representative augmentations simulating production conditions.
• To train a robust classifier using a custom lightweight convolutional neural network.

Main Methods:

• Data augmentation techniques were employed to create a comprehensive dataset reflecting real-world production scenarios.
• A custom lightweight convolutional neural network architecture was designed and implemented.
• Several regularization strategies were applied to the neural network to enhance its performance and generalization.
• The model was trained and evaluated on the augmented dataset for micro-crack detection.

Main Results:

• The proposed automated system successfully detected micro-cracks on photovoltaic cell surfaces.
• The custom convolutional neural network achieved a high F1-score of 85% for defect classification.
• The regularization strategies contributed to improved model performance and robustness.

Conclusions:

• The developed automated system offers a reliable and efficient alternative to manual inspection for photovoltaic cell micro-crack detection.
• The use of modelled data augmentations and a lightweight CNN architecture demonstrates a viable approach for improving solar cell manufacturing quality control.
• This research contributes to advancing non-destructive testing methods in the renewable energy sector.