ELVORA AI – Intelligent Solar Cell Defect Inspection

• 

  Software Dashboard

ELVORA AI is an intelligent solar cell defect inspection platform developed to improve photovoltaic quality control using artificial intelligence, computer vision, and cloud computing. The project was inspired by the need for faster, more consistent, and interpretable electroluminescence (EL) image analysis.

OpenCV 5 serves as the primary computer vision engine, handling image quality assessment, preprocessing, YOLO26n ONNX inference, defect localization, and geometric analysis. The platform also integrates YOLO11, YOLO26, Faster R-CNN, RT-DETR, HOG-SVM, and an adaptive Random-Forest router. AWS Lambda and Amazon S3 enable cloud-based image processing and result storage.

Our evaluation included 19,150 test images, with the OpenCV 5 pipeline achieving 0.4406 mAP50–95 and 0.6655 mAP50. The adaptive router achieved 0.8933 ROC-AUC.

Our main challenges involved integrating different AI architectures, preserving image information, and connecting local processing with AWS. We learned that reliable AI requires accurate detection, reproducibility, and transparent results.

ELVORA AI aims to make photovoltaic inspection more accessible, reliable, and efficient while supporting human engineering decisions.

Built With

• amazon-cloudwatch
• amazon-ecr
• amazon-web-services
• aws-iam
• aws-lambda
• boto3
• cnn
• faster-r-cnn
• hog
• numpy
• onnx
• opencv-5
• pyside6
• python
• pytorch
• random-forest
• rt-detr
• scikit-learn
• svm
• ultralytics
• yolo