Deteksi Arc Fault pada Sistem Kelistrikan Jaringan Tegangan Rendah Menggunakan CNN ResNet-50 Berbasis Mikroprosesor dengan Analisis Sinyal Arus Multidomain

Rizky, Ahmad (2026) Deteksi Arc Fault pada Sistem Kelistrikan Jaringan Tegangan Rendah Menggunakan CNN ResNet-50 Berbasis Mikroprosesor dengan Analisis Sinyal Arus Multidomain. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5022221192-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Gangguan busur listrik (arcing fault) tipe seri sangat sulit dideteksi oleh pengaman konvensional dan rentan memicu alarm palsu (nuisance tripping) akibat distorsi harmonisa perangkat elektronik. Untuk mengatasi hal ini, penelitian ini mengusulkan sistem proteksi cerdas waktu-nyata berbasis Edge AI pada platform NVIDIA Jetson Nano. Mengingat terbatasnya sumber daya komputasi tepi, sistem dilengkapi algoritma Gatekeeper yang secara asinkron menyaring beban normal, sehingga model AI hanya aktif saat terdeteksi anomali. Sinyal arus 1D kemudian ditransformasikan menjadi citra spektrogram 2D menggunakan Short-Time Fourier Transform (STFT) untuk diklasifikasikan oleh arsitektur Deep Residual Network (ResNet-50). Guna memenuhi standar waktu pemutusan keselamatan kelistrikan, eksekusi model diakselerasi menggunakan mesin NVIDIA TensorRT (reduksi presisi FP16). Hasil pengujian empiris membuktikan bahwa sistem bersifat amplitude-invariant, mampu mencapai akurasi klasifikasi 94,4% hingga 100% pada rentang variasi beban ekstrem (400 W hingga 1300 W). Sistem sukses mempertahankan tingkat alarm palsu 0% pada kondisi beban puncak, dengan latensi inferensi inti hanya sebesar 5,26 ms. Solusi Edge AI terintegrasi ini terbukti presisi, sangat efisien, kebal terhadap intervensi beban non-linier, dan siap untuk diimplementasikan langsung di lapangan.
Kata Kunci: Arc Fault, Edge AI, ResNet-50, Short-Time Fourier Transform (STFT), Nuisance Tripping, Beban Non-linier.
=======================================================================================================================================
Series arc faults are notoriously difficult to detect using conventional protective devices and are highly susceptible to triggering false alarms (nuisance tripping) due to harmonic distortions from electronic devices. To address this issue, this research proposes a real-time smart protection system based on Edge AI deployed on the NVIDIA Jetson Nano platform. Given the constrained resources of edge computing, the system features a Gatekeeper algorithm that asynchronously filters normal loads, ensuring the AI model is only activated upon detecting an anomaly. The 1D current signal is subsequently transformed into a 2D spectrogram image using the Short-Time Fourier Transform (STFT) for classification by a Deep Residual Network (ResNet-50) architecture. To meet strict electrical safety clearing time standards, model execution is accelerated using the NVIDIA TensorRT engine via FP16 precision reduction. Empirical evaluation results prove that the system is amplitude-invariant, achieving classification accuracies ranging from 94.4% to 100% across extreme load variations (400 W to 1300 W). The system successfully maintained a 0% false positive rate under peak load conditions, with a core inference latency of merely 5.26 ms. This integrated Edge AI solution has proven to be highly precise, efficient, immune to non-linear load interference, and ready for direct field deployment.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Arc Fault, Edge AI, ResNet-50, Short-Time Fourier Transform (STFT), Nuisance Tripping, Beban Non-linier, Arc Fault, Edge AI, ResNet-50, Short-Time Fourier Transform (STFT), Nuisance Tripping, Non-linear Load.
Subjects:	A General Works > AC Collections. Series. Collected works A General Works > AC Collections. Series. Collected works T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5102.9 Signal processing.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20201-(S1) Undergraduate Thesis
Depositing User:	Ahmad Rizky
Date Deposited:	24 Jul 2026 03:17
Last Modified:	24 Jul 2026 03:17
URI:	http://repository.its.ac.id/id/eprint/136723

Actions (login required)

View Item