Sistem Deteksi Kondisi Label Produk Pada Auto-Sorting System Menggunakan Convolutional Neural Network

Anjati, Anina (2024) Sistem Deteksi Kondisi Label Produk Pada Auto-Sorting System Menggunakan Convolutional Neural Network. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 2040201113-Undergraduate_Thesis for D4.pdf - Accepted Version Restricted to Repository staff only until 1 October 2026. Download (10MB) | Request a copy

Abstract

Label produk merupakan elemen penting dalam proses produksi dan distribusi yang berfungsi sebagai penanda informasi penting mengenai produk seperti identitas, kualitas, dan ketentuan lain dari sebuah produk. Keberhasilan deteksi kondisi label produk sangat penting untuk memastikan kualitas dan integritas produk yang dikirimkan kepada konsumen. Penelitian ini membahas implementasi metode Convolutional Neural Network (CNN) dalam sistem inspeksi visual sebagai solusi untuk mendapatkan nilai akurasi yang baik pada hasil deteksi kondisi label produk. Sistem inspeksi visual yang dikembangakan, terintegrasi dengan sebuah mesin sortir pada salah satu perusahaan manufaktur yang berfokus pada sektor otomasi yaitu Auto-Sorting System based Inspection Robot (ARSIR). Implementasi metode CNN diusulkan menjadi solusi atas keterbatasan ARSIR dalam mendeteksi kondisi label dengan posisi dan orientasi bervariasi diatas konveyor yang berada di luar jangkauan kamera. Keterbatasan ini dapat berpotensi meningkatkan kesalahan dalam pengambilan keputusan inspeksi. Penelitian dilakukan dengan membandingkan hasil model dari arsitektur SSD MobileNet-V2 dan EfficientDet-D0 dengan konfigurasi pelatihan batch size yang berbeda. Dataset yang digunakan berjumlah 1457 citra hasil augmentasi sebanyak tiga kali dengan menerapkan random rotation, brightness adjustment, dan gaussian blur. Dataset tersebut kemudian dilakukan pelabelan untuk mengklasifikasikan kondisi label produk. Hasil model terbaik didapatkan oleh arsitektur SSD MobileNet-V2 yang selanjutnya diimlementasikan pada sistem deteksi kondisi label produk. Metode CNN yang telah diimplementasikan mampu mendeteksi kondisi objek secara akurat dengan berbagai orientasi dan mampu memberikan respon sortir yang tepat untuk label produk yang sesuai dan tidak sesuai standar. Penelitian telah berhasil membuktikan bahwa model mampu mendeteksi kondisi label produk dengan berbagai orientasi dengan performa rata-rata baik dengan akurasi 99.26%, dengan precision 97.77%, recall 97.77% dan menghasilkan F1-Score 97,76%Penelitian ini telah memberikan nilai kualitas label produk yang sesuai standar pada proses inspeksi, meningkatkan kontrol kualitas, dan mengurangi atau menghilangkan potensi pekerjaan ulang (rework).
=================================================================================================================================
Product labels are crucial elements in the production and distribution process, serving as markers of important information about the product, such as identity, quality, and other specifications. The successful detection of product label conditions is essential to ensure the quality and integrity of products delivered to consumers. This study discusses the implementation of the Convolutional Neural Network (CNN) method in a visual inspection system as a solution to achieve high accuracy in detecting product label conditions. The developed visual inspection system is integrated with a sorting machine in a manufacturing company focused on automation, specifically the Auto-Sorting System based Inspection Robot (ARSIR). The implementation of the CNN method is proposed to address ARSIR's limitations in detecting label conditions with varying positions and orientations on a conveyor that are out of the camera's range. These limitations could potentially increase errors in inspection decision-making. The study was conducted by comparing model results from the SSD MobileNet-V2 and EfficientDet-D0 architectures with different batch size training configurations. The dataset used consisted of 1,457 images augmented three times by applying random rotation, brightness adjustment, and Gaussian blur. The dataset was then labeled to classify product label conditions. The best model results were obtained from the SSD MobileNet-V2 architecture, which was subsequently implemented in the product label condition detection system. The implemented CNN method was able to accurately detect object conditions with various orientations and provide correct sorting responses for labels that met and did not meet standards. The study successfully demonstrated that the model could detect product label conditions with various orientations, achieving an average performance with 99.26% accuracy, 97.77% precision, 97.77% recall, and an F1-Score of 97.76%. This research has provided product label quality standards in the inspection process, improved quality control, and reduced or eliminated the potential for rework.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Convolutional Neural Network, Deteksi Kondisi Label Produk, Inspeksi Visual, Sistem Sortir Otomatis, Automatic Sorter System, Product Label Condition Detection, Visual Inspection.
Subjects:	T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T59.7 Human-machine systems. T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5105.546 Computer algorithms T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7882.P3 Pattern recognition systems
Divisions:	Faculty of Vocational > 36304-Automation Electronic Engineering
Depositing User:	Anina Anjati
Date Deposited:	16 Aug 2024 01:12
Last Modified:	16 Aug 2024 01:12
URI:	http://repository.its.ac.id/id/eprint/115431

Actions (login required)

View Item