RANCANG BANGUN SISTEM IDENTIFIKASI PRODUK CACAT PADA FOLDING BOX BERBASIS YOLOv8m YANG TERINTEGRASI DENGAN REJECTION AREA

Agung Satria Pamungkas, - and Dewi Indriati Hadi Putri, - and Diky Zakaria, - (2025) RANCANG BANGUN SISTEM IDENTIFIKASI PRODUK CACAT PADA FOLDING BOX BERBASIS YOLOv8m YANG TERINTEGRASI DENGAN REJECTION AREA. S1 thesis, Universitas Pendidikan Indonesia.

Abstract

Industri manufaktur kemasan memerlukan sistem quality control yang cepat, akurat, dan efektif untuk menjaga kualitas produk. Penelitian ini merancang dan mengintegrasikan prototipe sistem produksi dengan algoritma deteksi cacat berbasis YOLOv8m. Sistem terdiri dari enam area: folding box placement, image acquisition, rejection, good folding box, computerization, dan air compression. Dataset tiga kategori (normal, robek, terlipat) diperoleh dari pengambilan gambar pada variasi ketinggian kamera, dianotasi menggunakan instance segmentation, dan diperluas dengan data augmentation. Model dilatih 50 epoch di Google Colab dan dievaluasi dengan akurasi, presisi, recall, mAP, dan F1-score. Metode dengan augmentation mencapai akurasi 0,906, presisi 0,964, recall 0,962, dan mAP 0,977, melampaui metode tanpa augmentation. Integrasi dengan sistem rejection memungkinkan deteksi dan pemisahan folding box cacat secara real-time tanpa intervensi manual, dengan FAR dan FRR rendah. Sistem stabil pada ketinggian kamera 30 cm maupun 40 cm dengan jarak antar folding box 20–35 cm. Penelitian ini menyimpulkan bahwa kombinasi hardware terintegrasi dan YOLOv8m menjadi solusi efektif untuk otomasi identifikasi produk cacat pada folding box di industri kemasan. ----- The packaging manufacturing industry requires a fast, accurate, and effective quality control system to maintain product standards. This study designed and integrated a production system prototype with a defect detection algorithm based on YOLOv8m. The system comprises six areas: folding box placement, image acquisition, rejection, good folding box, computerization, and air compression. A three-category dataset (normal, torn, folded) was obtained through image capture at varying camera heights, annotated using instance segmentation, and expanded via data augmentation. The model was trained for 50 epochs in Google Colab and evaluated using accuracy, precision, recall, mAP, and F1-score. The augmented method achieved 0.906 accuracy, 0.964 precision, 0.962 recall, and 0.977 mAP, outperforming the non-augmented method. Integration with the rejection system enabled real-time detection and removal of defective folding boxes without manual intervention, with low FAR and FRR. The system was stable at camera heights of 30 cm and 40 cm with box spacing of 20–35 cm. This study concludes that the combination of integrated hardware and YOLOv8m provides an effective solution for automating defect identification of folding boxes in the packaging industry.

	Text S_MKB_2104773_Title.pdf Download (528kB)
	Text S_MKB_2104773_Chapter1.pdf Download (593kB)
	Text S_MKB_2104773_Chapter2.pdf Restricted to Staf Perpustakaan Download (1MB)
	Text S_MKB_2104773_Chapter3.pdf Download (1MB)
	Text S_MKB_2104773_Chapter4.pdf Restricted to Staf Perpustakaan Download (2MB)
	Text S_MKB_2104773_Chapter5.pdf Download (302kB)
	Text S_MKB_2104773_Appendix.pdf Restricted to Staf Perpustakaan Download (856kB)

Item Type:	Thesis (S1)
Additional Information:	https://scholar.google.com/citations?hl=en&user=H0sTDGgAAAAJ ID SINTA Dosen Pembimbing: Dewi Indriati Hadi Putri: 6720737 Diky Zakaria: 6779007
Uncontrolled Keywords:	YOLOv8m, quality control, deteksi cacat, folding box, instance segmentation. YOLOv8m, quality control, defect detection, folding box, instance segmentation.
Subjects:	T Technology > T Technology (General) T Technology > TJ Mechanical engineering and machinery T Technology > TS Manufactures
Divisions:	UPI Kampus Purwakarta > S1 Mekatronika dan Kecerdasan Buatan
Depositing User:	Agung Satria Pamungkas
Date Deposited:	08 Sep 2025 08:11
Last Modified:	08 Sep 2025 08:11
URI:	http://repository.upi.edu/id/eprint/138155

Actions (login required)

View Item