Analisis Komparatif Model CNN Googlenet, Vgg16, Dan Resnet50 Terhadap Inspeksi Manual Dalam Optimalisasi Quality Control Roti Bluder Pada Lini Produksi

Wulandari, Esty (2025) Analisis Komparatif Model CNN Googlenet, Vgg16, Dan Resnet50 Terhadap Inspeksi Manual Dalam Optimalisasi Quality Control Roti Bluder Pada Lini Produksi. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6007231006-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Proses quality control (QC) pada industri roti, khususnya roti bluder, selama ini masih dilakukan secara manual dan memiliki beberapa keterbatasan, seperti ketergantungan pada ketelitian individu, ketidakkonsistenan hasil, serta risiko terlewatnya produk cacat yang dapat berdampak pada keluhan pelanggan dan citra perusahaan. Penelitian ini dilakukan untuk mengkaji dan membandingkan performa tiga model Convolutional Neural Network (CNN), yaitu GoogLeNet, VGG16, dan ResNet50, dalam mendeteksi cacat visual roti bluder, serta membandingkannya dengan hasil inspeksi manual yang dilakukan oleh karyawan berdasarkan pengalaman kerja.
Data citra dikumpulkan langsung dari lini produksi dan diklasifikasikan ke dalam tiga kategori: produk sesuai spesifikasi, cacat karena cemaran/pengotor, dan cacat karena isian meluber. Selanjutnya dari data citra disusun pengaturan parameter pada model CNN GoogLeNet, VGG16, dan ResNet50. Model dilatih dan diuji menggunakan perangkat lunak MATLAB dengan evaluasi berdasarkan metrik akurasi, presisi, dan recall serta performance dari model. Hasil penelitian menunjukkan bahwa model CNN mampu memberikan performa deteksi yang lebih tinggi dibandingkan tenaga manusia. ResNet50 dan GoogLeNet memberikan akurasi pengujian sebesar 100%, disusul VGG16 sebesar 96,67%. Sementara itu, inspeksi manual oleh karyawan menghasilkan akurasi yang jauh lebih rendah, yaitu sebesar 72,35% untuk karyawan dengan pengalaman kerja 12–60 bulan, dan 70,46 % untuk karyawan dengan pengalaman lebih dari 60 bulan. Hasil dari penelitian ini menunjukkan bahwa implementasi control kualitas berbasis CNN tidak hanya mampu meningkatkan akurasi dan konsistensi proses QC, tetapi juga dapat mengurangi potensi kerugian akibat produk cacat yang lolos sortir. Oleh karena itu, implementasi kontrol kualitas berbasis CNN sangat direkomendasikan sebagai solusi untuk mengoptimalkan pengendalian mutu produk di industri pangan.
========================================================================================================================================
The quality control (QC) process in the bread industry, particularly for bluder bread, has traditionally been conducted manually and is subject to several limitations, such as dependence on individual accuracy, inconsistency in results, and the risk of defective products being overlooked. These issues can lead to customer complaints and damage the company’s reputation. This study was conducted to evaluate and compare the performance of three Convolutional Neural Network (CNN) models—GoogLeNet, VGG16, and ResNet50—in detecting visual defects in bluder bread, and to benchmark their performance against manual inspection carried out by employees based on their work experience.Image data were collected directly from the production line and classified into three categories: products meeting specifications, defects caused by contamination/foreign matter, and defects caused by filling overflow. These images were used to configure parameters for the GoogLeNet, VGG16, and ResNet50 CNN models. The models were trained and tested using MATLAB software and evaluated using accuracy, precision, and recall metrics, along with overall model performance.The results demonstrated that CNN models provided significantly higher detection performance compared to human inspectors. Both ResNet50 and GoogLeNet achieved a test accuracy of 100%, while VGG16 followed with 96.67%. In contrast, manual inspection by employees resulted in substantially lower accuracy, with 72.35% for employees with 12–60 months of experience and 70.46% for those with over 60 months of experience. These findings suggest that implementing CNN-based quality control not only enhances the accuracy and consistency of QC processes but also reduces the potential losses caused by defective products passing inspection. Therefore, CNN-based quality control systems are strongly recommended as a solution to optimize product quality assurance in the food industry.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Convolutional Neural Network, Quality Control, Roti Bluder, Industri Pangan, VGG16, GoogLeNet, ResNet50,Quality Control, Bluder Bread, Food Industry
Subjects:	T Technology > TS Manufactures T Technology > TS Manufactures > TS155 Production control. Production planning. Production management T Technology > TS Manufactures > TS176 Manufacturing engineering. Process engineering (Including manufacturing planning, production planning)
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Mechanical Engineering > 21101-(S2) Master Thesis
Depositing User:	Esty Wulandari
Date Deposited:	04 Aug 2025 05:25
Last Modified:	04 Aug 2025 08:48
URI:	http://repository.its.ac.id/id/eprint/126410

Actions (login required)

View Item