Deteksi Dini Api Menggunakan Sensor Gas dan Fotodetektor Berbasis Machine Learning

Darmawan, Andi (2026) Deteksi Dini Api Menggunakan Sensor Gas dan Fotodetektor Berbasis Machine Learning. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5022221024-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (6MB) | Request a copy

Abstract

Kebakaran akibat korsleting listrik masih menjadi salah satu penyebab utama kebakaran di lingkungan perkotaan. Sistem deteksi konvensional umumnya memberikan peringatan setelah muncul asap atau api, sehingga diperlukan metode deteksi yang lebih dini. Penelitian ini mengembangkan sistem deteksi dini kebakaran berbasis deteksi busur listrik menggunakan sensor UVTRON R2868 dan sensor gas MQ131 yang dipadukan dengan algoritma machine learning. Algoritma yang diuji meliputi Random Forest, AdaBoost, Gradient Boosting, dan Long Short-Term Memory (LSTM). Hasil pengujian offline menunjukkan LSTM memberikan performa terbaik dengan akurasi 97,56% dan recall 100%. Pengujian online selama 8 jam menunjukkan hanya Gradient Boosting yang menghasilkan satu false positive, sedangkan model lainnya stabil. Pada pengujian kondisi aktif, seluruh model mampu mendeteksi kejadian busur listrik tanpa false negative. Hasil penelitian menunjukkan bahwa sensor fusion meningkatkan performa deteksi dibandingkan sensor tunggal. Meskipun LSTM memiliki performa terbaik, Random Forest Embedded dipilih untuk implementasi pada sistem embedded karena memiliki kinerja yang kompetitif (akurasi 93,67%) dan dapat dijalankan pada mikrokontroler STM32F103C8T6 dengan keterbatasan sumber daya.
=====================================================================================================================================
Electrical short-circuit-induced fires remain one of the leading causes of urban fire incidents. Conventional fire detection systems generally issue warnings only after smoke or flames have appeared, highlighting the need for earlier detection methods. This study develops an early fire detection system based on electric arc detection using a UVTRON R2868 sensor and an MQ131 gas sensor combined with machine learning algorithms. The evaluated algorithms include Random Forest, AdaBoost, Gradient Boosting, and Long Short-Term Memory (LSTM). Offline testing results show that the LSTM model achieved the best performance, with an accuracy of 97.56% and a recall of 100%. During an 8-hour online evaluation, only the Gradient Boosting model produced one false positive, while the remaining models operated stably. Under active testing conditions, all models successfully detected electric arc events without any false negatives. The results demonstrate that sensor fusion improves detection performance compared with the use of a single sensor. Although LSTM achieved the highest overall performance, the embedded Random Forest model was selected for deployment due to its competitive performance (93.67% accuracy) and its ability to operate efficiently on the resource-constrained STM32F103C8T6 microcontroller.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Busur Listrik, Deteksi Dini Kebakaran, Machine Learning, MQ131, UVTRON, Arc Fault, Early Fire Detection, Machine Learning, MQ131, UVTRON.
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK152.A75 Electrical engineering--Safety measures T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5105.546 Computer algorithms T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7871.674 Detectors. Sensors T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7882.P3 Pattern recognition systems
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20201-(S1) Undergraduate Thesis
Depositing User:	Andi Darmawan
Date Deposited:	23 Jul 2026 08:35
Last Modified:	23 Jul 2026 08:35
URI:	http://repository.its.ac.id/id/eprint/136588

Actions (login required)

View Item