Optimalisasi Deteksi Kerusakan Elektrikal Panel Surya dengan Transfer Learning dan Augmentasi Terkontrol berbasis YOLOv8

Abstrak

Electrical fault detection in solar panels is a critical challenge in maintaining the efficiency of large-scale photovoltaic energy systems. This research develops a deep learning-based automated classification model by leveraging the YOLOv8-CLS architecture, refined through transfer learning and systematically applied data augmentation. The dataset consists of two panel condition classes, clean and electrical-damage, which were preprocessed through image size normalization, tensor transformation, and augmentation using RandAugment and random erasing. The model was trained for 15 epochs with fine-tuning applied to the head, while the backbone retained pretrained weights. Performance evaluation showed that the model achieved a Top-1 Accuracy of 98.21%, with precision for the electrical-damage class reaching 100%, recall at 94.12%, and an F₁-score of 0.9697. Furthermore, an average inference time of 18.82 milliseconds per image demonstrates high computational efficiency for real-time deployment. These findings indicate that the integration of the YOLOv8 architecture with transfer learning and controlled augmentation is effective for detecting electrical faults in solar panels and is suitable for implementation in automated monitoring systems based on edge or cloud computing.