An ML framework detects faults in PV panels from their unlabeled EL images using transfer learning, k -means labeling, and data augmentation. CNN models achieve 98–99.5% accuracy for binary classification, reducing manual labeling and improving PV system reliability.