Deteksi Alat Pelindung Diri Menggunakan Metode RT-DETR melalui Mannequin Safety Officer sebagai Sistem Pengawasan Keselamatan Kerja

Alfiansyah, Moch Rio (2026) Deteksi Alat Pelindung Diri Menggunakan Metode RT-DETR melalui Mannequin Safety Officer sebagai Sistem Pengawasan Keselamatan Kerja. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 2040221148_Undergraduate_thesis.pdf Restricted to Repository staff only Download (7MB) | Request a copy

Abstract

Compliance with Personal Protective Equipment (PPE) requirements is an important aspect of Occupational Health and Safety (OHS) implementation in the Warehousing and Bagging Division of PT Petrokimia Gresik. Although PPE regulations have been implemented, noncompliance is still found among workers and other individuals entering operational areas. Based on OHS inspection data, 11 violations were recorded over one month at several operational locations, while a five-hour direct observation at the Multipurpose Warehouse identified 27 PPE violations. Manual supervision has limitations because officers cannot always be present at every location when a violation occurs. In addition, supervision may be affected by workload, competing priorities, and fatigue resulting from continuous monitoring. This study aims to develop a Mannequin Safety Officer as a supporting monitoring system capable of automatically detecting PPE violations in real time within areas covered by the camera, providing immediate audio warnings, and recording detection results as supporting data for OHS monitoring and evaluation. The initial dataset consisted of 4,149 images collected from several operational areas under daytime and nighttime conditions. After augmentation was applied to the training data, the dataset consisted of 6,078 training images and 1,110 validation images with seven classes: person, helmet, seragam_dinas, boots, no_helmet, no_seragam_dinas, and no_boot. The RT-DETR model was trained for 70 epochs and achieved a precision of 0.9559, a recall of 0.9560, an mAP@0.50 of 0.9773, and an mAP@0.50:0.95 of 0.7299. The test results showed that the system successfully activated audio warnings when PPE violations were detected and remained inactive when complete PPE was worn. Detection results were automatically recorded in Excel files, accompanied by image evidence, and stored in Google Drive. During battery testing, the system was able to operate for one day, 14 hours, 52 minutes, and 41 seconds. Therefore, the developed system can assist officers in detecting, warning, and recording PPE violations at monitoring points within the camera’s coverage while providing supporting data for OHS evaluation.
==================================================================================================================================
Kepatuhan penggunaan Alat Pelindung Diri (APD) merupakan bagian penting dalam penerapan Keselamatan dan Kesehatan Kerja (K3) pada area Pergudangan dan Pengantongan PT Petrokimia Gresik. Meskipun ketentuan penggunaan APD telah diberlakukan, masih ditemukan ketidakpatuhan penggunaan APD oleh pekerja maupun pihak lain yang memasuki area operasional. Berdasarkan data temuan K3, tercatat 11 pelanggaran selama satu bulan pada beberapa lokasi operasional, sedangkan observasi langsung selama lima jam di Gudang Multiguna menemukan 27 pelanggaran penggunaan APD. Pengawasan manual memiliki keterbatasan karena petugas tidak selalu berada pada setiap titik ketika pelanggaran terjadi. Selain itu, pengawasan dapat dipengaruhi oleh beban kerja, prioritas terhadap aktivitas lain, serta kelelahan akibat pemantauan secara terus-menerus. Penelitian ini bertujuan mengembangkan Mannequin Safety Officer sebagai sistem pendukung pengawasan yang mampu mendeteksi pelanggaran APD secara otomatis pada area yang berada dalam jangkauan kamera, memberikan peringatan suara secara langsung, serta mencatat hasil deteksi sebagai data pendukung pemantauan dan evaluasi K3. Dataset awal terdiri atas 4.149 citra yang diperoleh dari beberapa area operasional pada kondisi siang dan malam. Setelah augmentasi diterapkan data latih, dataset terdiri atas 6.078 citra latih dan 1.110 citra validasi dengan tujuh kelas, yaitu person, helmet, seragam_dinas, boots, no_helmet, no_seragam_dinas, dan no_boot. Model RT-DETR dilatih selama 70 epoch dan menghasilkan nilai precision sebesar 0,9559, recall sebesar 0,9560, mAP@0,50 sebesar 0,9773, serta mAP@0,50:0,95 sebesar 0,7299. Hasil pengujian menunjukkan sistem mampu mengaktifkan peringatan suara ketika pelanggaran APD terdeteksi dan tidak mengaktifkan peringatan ketika APD digunakan secara lengkap. Hasil deteksi dapat dicatat secara otomatis dalam berkas Excel, disertai bukti citra, dan disimpan pada Google Drive. Pada pengujian baterai, sistem mampu beroperasi selama 1 hari, 14 jam, 52 menit, dan 41 detik. Sistem yang dikembangkan dapat membantu petugas dalam mendeteksi, memberikan peringatan, dan mencatat pelanggaran penggunaan APD pada titik pemantauan yang berada dalam jangkauan kamera serta menyediakan data pendukung evaluasi K3.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Alat Pelindung Diri, RT-DETR, computer vision, edge computing, keselamatan kerja Personal Protective Equipment, RT-DETR, computer vision, edge computing, occupational safety
Subjects:	Q Science > QA Mathematics > QA76.758 Software engineering Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) T Technology > T Technology (General) > T55 Industrial Safety
Divisions:	Faculty of Vocational > 36304-Automation Electronic Engineering
Depositing User:	Moch. Rio Alfiansyah
Date Deposited:	07 Aug 2026 02:06
Last Modified:	07 Aug 2026 02:06
URI:	http://repository.its.ac.id/id/eprint/142993

Actions (login required)

View Item