DETEKSI KERUSAKAN PADA PANEL SURYA MENGGUNAKAN METODE YOU ONLY LOOK ONCE (YOLO) v8

Rabbani, Naufal Irbah and Cahyaningtyas, Rizqia and Sudirman, M. Yoga Distra (2025) DETEKSI KERUSAKAN PADA PANEL SURYA MENGGUNAKAN METODE YOU ONLY LOOK ONCE (YOLO) v8. Diploma thesis, ITPLN.

[thumbnail of 202131192_Naufal Irbah Rabbani_Revisi_Skripsi_Naufal Irbah Rabbani.pdf] Text
202131192_Naufal Irbah Rabbani_Revisi_Skripsi_Naufal Irbah Rabbani.pdf
Restricted to Registered users only

Download (8MB)

Abstract

Pemeliharaan panel surya merupakan tantangan utama dalam industri energi terbarukan, di mana inspeksi manual bersifat tidak efisien, memakan waktu, dan berisiko melewatkan kerusakan yang dapat menurunkan produktivitas sistem. Penelitian ini bertujuan untuk mengembangkan dan mengevaluasi sistem deteksi kerusakan otomatis pada panel surya menggunakan model You Only Look Once versi 8 (YOLOv8). Model ini dilatih pada dataset citra untuk mengidentifikasi kerusakan spesifik seperti Crack (retakan), Burn (terbakar), dan Snail Track (jalur siput), serta panel dalam kondisi Non-Defective. Evaluasi awal menggunakan confusion matrix menunjukkan model YOLOv8m dengan performa terbaik (F1-score 58%) dan YOLOv8l dengan hasil ter rendah (F1-score 56%). Model YOLOv8l kemudian dipilih untuk validasi lanjutan menggunakan K-Fold Cross Validation guna meningkatkan kepercayaan dan mengukur batas minimal berforma model. Proses validasi ini terbukti sukses, di mana pada fold keempat, model mencapai peningkatan performa yang signifikan dengan F1-score 74,2% dan mAP 77,6%. Kinerja tertinggi tercatat pada deteksi kelas Non-Defective dan Snail Track, sedangkan kelas Crack merupakan kategori yang paling sulit untuk dideteksi. Hasil ini membuktikan bahwa sistem berbasis YOLOv8 dengan validasi K-Fold merupakan solusi andal untuk pemantauan otomatis, yang dapat meningkatkan efisiensi pemeliharaan dan mengurangi biaya operasional di industri energi surya.

Solar panel maintenance is a major challenge in the renewable energy industry, where manual inspection is inefficient, time-consuming, and risks missing defects that can reduce system productivity. This study aims to develop and evaluate an automated damage detection system for solar panels using the You Only Look Once version 8 (YOLOv8) model. This model was trained on an image dataset to identify specific defects such as Cracks, Burns, and Snail Tracks, as well as panels in Non-Defective condition. Initial evaluation using a confusion matrix showed the YOLOv8m model with the best performance (F1-score 58%) and YOLOv8l with the lowest results (F1-score 56%). The YOLOv8l model was then selected for further validation using K-Fold Cross Validation to increase confidence and measure the minimum performance threshold of the model. This validation process proved successful, where in the fourth fold, the model achieved a significant performance improvement with an F1-score of 74.2% and mAP of 77.6%. The highest performance was recorded in the Non-Defective and Snail Track detection classes, while the Crack class was the most difficult to detect. These results demonstrate that the YOLOv8-based system with K-Fold validation is a reliable solution for automated monitoring, which can improve maintenance efficiency and reduce operational costs in the solar energy industry..

Item Type: Thesis (Diploma)
Uncontrolled Keywords: YOLOv8, Deteksi Objek, Panel Surya, Kerusakan Panel Surya, Deep Learning, K-Fold Cross Validation, Computer Vision. YOLOv8, Object Detection, Solar Panel, Solar Panel Defect, Deep Learning, K-Fold Cross Validation, Computer Vision.
Subjects: Skripsi
Bidang Keilmuan > Teknik Informatika
Divisions: Fakultas Telematika Energi > S1 Teknik Informatika
Depositing User: Sudarman
Date Deposited: 14 Oct 2025 08:22
Last Modified: 14 Oct 2025 08:22
URI: https://repository.itpln.ac.id/id/eprint/2300

Actions (login required)

View Item
View Item