Sistem Inspeksi Visual Untuk Kondisi Kelengkapan Isi Folding Box Pada Proses Angkat Obat Dengan Metode Convolutional Neural Network (CNN)

Rachiemullah, Ilham Akbar (2025) Sistem Inspeksi Visual Untuk Kondisi Kelengkapan Isi Folding Box Pada Proses Angkat Obat Dengan Metode Convolutional Neural Network (CNN). Diploma thesis, Institut Teknologi Sepuluh Nopember.

Text 2040211055-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (6MB) | Request a copy

Abstract

Proses pengemasan di industri seringkali menghadapi masalah ketidaklengkapan isi "folding box", terutama pada tahap angkat obat yang masih mengandalkan tenaga manusia. Hal ini terbukti dari data September 2023 hingga September 2024, di mana terjadi 16,6% kasus isi kardus kurang akibat error mesin counting atau kelalaian pergantian shift. Untuk meminimalisir masalah ini, penelitian ini mengembangkan sistem inspeksi visual otomatis menggunakan metode Convolutional Neural Network (CNN) dengan algoritma You Only Look Once (YOLO). Sistem ini terintegrasi dengan Mini PC Nvidia Jetson Orin Nano, kamera, dan buzzer. Sistem bekerja kerjanya melibatkan pengambilan citra folding box di dalam kardus oleh kamera, kemudian dianalisis oleh Mini PC untuk klasifikasi kondisi: "Isi-Penuh", "Kurang-10-Folding-Box", "Kurang-20-Folding-Box", dan "Kurang-30-Folding-Box". Jika terdeteksi kondisi "kurang", buzzer akan aktif sebagai peringatan dini bagi operator. Hasil pengujian menunjukkan efektivitas dan keandalan sistem. Model YOLOv5s dan YOLOv5m memiliki kinerja lokalisasi bounding box dan klasifikasi objek yang lebih stabil dan unggul. Meskipun semua model YOLO mencapai metrik Precision, Recall, dan mAP tinggi (di atas 90-95%), strategi anotasi sangat memengaruhi performa; Anotasi Volume 1 unggul dalam mendeteksi "Isi-Penuh" (99%) dan "Kurang-30-Folding-Box" (100%), sementara Anotasi Volume 2 menunjukkan penurunan signifikan pada "Kurang-20-Folding-Box" namun sangat baik untuk "Kurang-10-Folding-Box". Penempatan kamera optimal adalah 90 cm, mencapai recall 100% untuk semua kelas. Secara keseluruhan, sistem ini berpotensi besar meminimalkan isu isi kardus kurang dan meningkatkan efisiensi proses pengemasan
=====================================================================================================================================
Industrial packaging processes often face issues with incomplete "folding box" contents, particularly in the drug lifting stage, which still relies on human labor. This is evidenced by data from September 2023 to September 2024, showing 16.6% of cases with insufficient box contents due to counting machine errors or oversight during shift changes. To mitigate this problem, this research develops an automatic visual inspection system using the Convolutional Neural Network (CNN) method with the You Only Look Once (YOLO) algorithm. The system integrates an Nvidia Jetson Orin Nano Mini PC, a camera, and a buzzer. Its operation involves the camera capturing images of the folding boxes inside the cartons, which are then analyzed by the Mini PC for classification into four conditions: "Full-Content”, " Less-10-Boxes, "Less-20-Boxes”, and "Less-30-Boxes”. If an "insufficient" condition is detected, the buzzer activates as an early warning for the operator. Test results demonstrate the system's effectiveness and reliability. YOLOv5s and YOLOv5m models showed superior and more stable performance in bounding box localization and object classification. Although all YOLO models (YOLOv8s, YOLOv8m, YOLOv5s, and YOLOv5m) achieved high precision, recall, and mAP metrics (often above 90-95%), the annotation strategy significantly impacted performance. Annotation Volume 1 excelled in detecting "Full-Content" (99%) and "Less-30-Boxes" (100%), while Annotation Volume 2 showed a significant drop in "Kurang-20-Folding-Box" performance (F1-score only 67.96%) but was very effective for "Less-30-Boxes" (F1-score 96.96%, recall 98.48%, precision 95.58%). Optimal camera placement was determined to be 90 cm, achieving a consistent 100% recall for all classes. Overall, this system holds significant potential to minimize issues of insufficient carton contents and enhance packaging process efficiency

Item Type:	Thesis (Diploma)
Uncontrolled Keywords:	Isi Kurang , Inspeksi Visual , Convolutional Neural Network (CNN), You Only Look Once (YOLO), Less Boxes, Visual Inspection, Convolutional Neural Network (CNN), You Only Look Once (YOLO)
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5102.9 Signal processing. T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7871.674 Detectors. Sensors T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7888.3 Digital computers
Divisions:	Faculty of Vocational > 36304-Automation Electronic Engineering
Depositing User:	Ilham Akbar Rachiemullah
Date Deposited:	13 Aug 2025 07:24
Last Modified:	13 Aug 2025 07:24
URI:	http://repository.its.ac.id/id/eprint/128085

Actions (login required)

View Item