Sistem Deteksi Benda Asing Pada Loader Mesin Penghancur Pupuk Dengan Metode Single Shot Multibox Detector (SSD)

Samuel, Theodore (2025) Sistem Deteksi Benda Asing Pada Loader Mesin Penghancur Pupuk Dengan Metode Single Shot Multibox Detector (SSD). Other thesis, Institut Teknologi Sepuluh Nopember.

Text 2040211036-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

Sistem penghancur pupuk sering menghadapi masalah kerusakan akibat masuknya benda asing seperti besi dan beton, yang menyebabkan downtime dan biaya perbaikan yang tinggi. Penelitian ini bertujuan mengembangkan sistem deteksi otomatis untuk mengatasi permasalahan tersebut, dengan fokus pada perbandingan metode Single Shot Multibox Detector (SSD) dan You Only Look Once (YOLO) untuk menentukan metode dengan akurasi tertinggi serta menganalisis model mana yang paling unggul untuk implementasi pada sistem terbatas yang hanya mengandalkan CPU (CPU-only). Metodologi yang diterapkan meliputi perancangan sistem berbasis kamera webcam yang memonitor konveyor, serta pelatihan dua model secara terpisah: SSD MobileNetV2 dan YOLOv8, menggunakan dataset yang dikumpulkan dari lingkungan industri. Dataset yang digunakan adalah benda asing yang bercampur dengan pupuk seperti beton, besi, dan rangka mesin lainya. Total jumlah dataset digunakan sebanyak 2143 gambar dan memiliki 3437 total label tag. Hasil pengujian secara keseluruhan menunjukkan bahwa meskipun model YOLOv8 memiliki tingkat akurasi (81,86%), presisi (100%), dan recall (69,29%) yang lebih tinggi, performanya sangat tidak stabil dan lambat (di bawah 5 FPS) saat diuji pada sistem CPU-only. Sebaliknya, model SSD terbukti jauh lebih ringan, stabil, dan mampu berjalan konsisten pada kecepatan 12-15 FPS dalam kondisi yang sama. Dengan demikian, disimpulkan bahwa SSD merupakan pilihan yang lebih superior dan realistis untuk implementasi di lapangan yang memiliki keterbatasan sumber daya komputasi.
============================================================================================================================================
The fertilizer crushing system often faces problems of damage due to the entry of foreign objects such as iron and concrete, which leads to downtime and high repair costs. This research aims to develop an automatic detection system to overcome this problem, focusing on a comparison of the Single Shot Multibox Detector (SSD) and You Only Look Once (YOLO) methods to determine the method with the highest accuracy and to analysed which model is most superior for implementation on a limited system that relies solely on a CPU (CPU-only). The methodology applied includes designing a webcam-based system that monitors the conveyor, as well as training two separate models: SSD MobileNetV2 and YOLOv8, using a dataset collected from an industrial environment. The dataset used consists of foreign objects mixed with fertilizer, such as concrete, iron, and other machine frames. The total dataset used comprises 2143 images and has a total of 3437 label tags. The overall test results show that although the YOLOv8 model has a higher accuracy (81.86%), precision (100%), and recall (69.29%), its performance is very unstable and slow (below 5 FPS) when tested on a CPU-only system. In contrast, the SSD model proved to be much lighter, more stable, and capable of running consistently at a speed of 12-15 FPS under the same conditions. Thus, it is concluded that SSD is a more superior and realistic choice for implementation in the field where computational resources are limited.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Crushing Machine, Early Warning System, Single Shot Multibox Detection
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7871.674 Detectors. Sensors
Divisions:	Faculty of Civil Engineering and Planning > Civil Engineering > 22301-(D4) Diploma 4
Depositing User:	Theodore Toby Samuel
Date Deposited:	01 Aug 2025 09:04
Last Modified:	01 Aug 2025 09:04
URI:	http://repository.its.ac.id/id/eprint/125590

Actions (login required)

View Item