Design of Defect Detection System with Real time Surveillance on Pilfer Proof Caps

Helisa, Yasmin Nur (2025) Design of Defect Detection System with Real time Surveillance on Pilfer Proof Caps. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5010211136-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Permintaan yang terus meningkat terhadap kemasan berkualitas tinggi mendorong kebutuhan akan sistem inspeksi otomatis yang andal dalam proses manufaktur. Metode inspeksi manual konvensional, meskipun masih umum digunakan, memiliki keterbatasan signifikan akibat kelelahan operator, subjektivitas penilaian, serta ketidaksesuaian dengan kecepatan produksi yang tinggi. Penelitian ini mengusulkan suatu sistem deteksi cacat secara waktu nyata yang mengintegrasikan metode deep transfer learning dengan sistem penglihatan multi-sudut berbasis kamera web untuk inspeksi kualitas pilfer-proof cap (PP) model YOLOv5 yang telah disempurnakan dengan penambahan Gradient Activation Module (GAM) dan Coordinate Spatial Information (CSI), dibandingkan secara komprehensif dengan model ResNet50, Faster R-CNN, dan MobileNetV2 yang telah dioptimalkan, menunjukkan kinerja unggul dalam hal akurasi, presisi, dan kecepatan inferensi. Sistem ini memanfaatkan tiga kamera web yang diposisikan pada sudut atas, depan, dan bawah untuk menangkap citra dari berbagai perspektif, yang kemudian ditampilkan secara langsung melalui antarmuka dashboard berbasis Flask. Mekanisme ambang batas kepercayaan berbasis Youden’s Index diterapkan untuk meningkatkan keandalan prediksi dengan meminimalkan tingkat false positive dan false negative. Variasi kondisi pencahayaan dikendalikan melalui penggunaan kotak kontrol tertutup guna menstabilkan input visual, sementara keberagaman dataset diperluas untuk meningkatkan generalisasi model. Hasil pengujian menunjukkan pengurangan waktu inspeksi sebesar 19,33% dibandingkan inspeksi manual, mengindikasikan kelayakan sistem untuk diterapkan dalam lingkungan industri. Solusi ini menawarkan pendekatan yang efektif dan skalabel untuk otomatisasi inspeksi kualitas pada lini produksi kemasan modern.
======================================================================================================================================
The increasing demand for high-quality packaging has intensified the need for reliable, automated inspection systems in manufacturing. Traditional manual inspections, though widely used, are hindered by operator fatigue, subjective errors, and their inability to meet high-speed production requirements. This study presents a real-time defect detection system that integrates deep transfer learning and a webcam-based multi-view vision setup for pilfer-proof (PP) cap quality inspection. An enhanced YOLOv5 model, augmented with a Gradient Activation Module (GAM) and Coordinate Spatial Information (CSI), is benchmarked against improved models of ResNet50, Faster R-CNN, and MobileNetV2, demonstrating superior accuracy, precision, and inference speed. The system captures top, front, and bottom views of PP caps using strategically positioned web cameras and delivers live defect detection via a Flask-based dashboard. A Youden's Index-based confidence threshold is implemented to enhance inference reliability by minimizing both false positives and false negatives. Environmental variability is addressed through an enclosed lighting control box to stabilize the camera input, while dataset diversity is emphasized to enhance model generalization. System testing shows a 19.33% reduction in inspection time compared to manual inspection, validating its feasibility for industrial deployment. The proposed solution provides a practical and scalable approach to automated quality inspection, enhancing production efficiency and compliance in modern packaging lines.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Deep Learning, Transfer Learning, Computer Vision, Defect Detection, Pilfer-Proof Caps, Industrial Automation, Deep Learning, Transfer Learning, Computer Vision, Deteksi Cacat, Pilfer-Proof Cap, Otomasi Industri
Subjects:	T Technology > TS Manufactures > TS156 Quality Control. QFD. Taguchi methods (Quality control)
Divisions:	Faculty of Industrial Technology > Industrial Engineering > 26201-(S1) Undergraduate Thesis
Depositing User:	Yasmin Nur Helisa
Date Deposited:	18 Jul 2025 07:17
Last Modified:	18 Jul 2025 07:17
URI:	http://repository.its.ac.id/id/eprint/120010

Actions (login required)

View Item