Deteksi Dini Kelelahan Otot Berbasis Fitur Time-Frequency Dari Surface EMG Menggunakan Deep Learning Pada Perangkat Edge

Pribadi, Muhammad Fauzan Nur Insan (2024) Deteksi Dini Kelelahan Otot Berbasis Fitur Time-Frequency Dari Surface EMG Menggunakan Deep Learning Pada Perangkat Edge. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5024201019-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2026. Download (7MB) | Request a copy

Abstract

Kelelahan otot merupakan kondisi umum yang dapat menyebabkan penurunan kinerja fisik dan meningkatkan risiko cedera. Deteksi kelelahan otot secara dini dapat membantu mencegah cedera dan meningkatkan keselamatan. Penelitian ini bertujuan untuk mengembangkan model deteksi kelelahan otot menggunakan sinyal EMG. Data sinyal EMG dikumpulkan dari 7 orang responden yang melakukan isometric exercise dengan tiga variasi berat dumbell. Data EMG difilter untuk menghilangkan noise dan motion artifacts, kemudian disegmentasi dan di-windowing. Fitur-fitur penting diekstraksi menggunakan metode ekstraksi fitur domain waktu dan frekuensi tradisional serta ekstraksi fitur dengan metode time-frequency seperti Short-Time Fourier Transform (STFT) dan Continuous Wavelet Transform (CWT), lalu dilabeli berdasarkan tingkat kelelahan otot menggunakan Zero Crossing Rate dan Median Frequency. Model deteksi dini kelelahan otot dikembangkan menggunakan Support Vector Machine (SVM) dan Random Forest (RF). Model akan diimplementasikan pada perangkat edge, yaitu Raspberry Pi. Hasil penelitian menunjukkan bahwa sinyal sEMG dari Myo Arm Band mengalami clipping, yang mempengaruhi pengolahan sinyal. Parameter tuning SVM menunjukkan bahwa parameter C yang optimal berkisar antara 10 hingga 100, dengan γ sebesar 1. Untuk Random Forest, parameter optimal meliputi jumlah estimator 100-200, max features ‘sqrt’, max depth 10, dan min samples split 2 atau 5. Set fitur tradisional mencapai akurasi tertinggi (84.21%), diikuti oleh Wavelet (78.07%) dan STFT (72.81%). Varian preprocessing data 0.0 memiliki akurasi tertinggi (75.16%), diikuti varian 4.0 (75%). Model Random Forest menunjukkan akurasi tertinggi namun cenderung overfitting, sementara SVM lebih seimbang dalam precision dan recall. SVM juga lebih cepat dalam pemrosesan dibandingkan Random Forest, dengan set fitur STFT memiliki waktu pemrosesan tercepat.
======================================================================================================
Muscle fatigue is a common condition that can lead to reduced physical performance and increased risk of injury. Early detection of muscle fatigue can help prevent injuries and improve safety. This study aims to develop a muscle fatigue detection model using EMG signals. EMG data were collected from 7 respondents performing isometric exercises with three different dumbbell weights. The EMG data were filtered to remove noise and motion artifacts, then segmented and windowed. Key features were extracted using traditional time and frequency domain methods as well as time-frequency methods like Short-Time Fourier Transform (STFT) and Continuous Wavelet Transform (CWT). The data were labeled based on muscle fatigue levels using Zero Crossing Rate and Median Frequency. The early muscle fatigue detection model was developed using Support Vector Machine (SVM) and Random Forest (RF), and implemented on an edge device, the Raspberry Pi. The study found that EMG signals from the Myo Arm Band experienced clipping, affecting signal processing. SVM parameter tuning showed that the optimal C parameter ranged from 10 to 100, with γ set to 1. For Random Forest, optimal parameters included 100-200 estimators, 'sqrt' for max features, a max depth of 10, and min samples split of 2 or 5. The traditional feature set achieved the highest accuracy (84.21%), followed by Wavelet (78.07%) and STFT (72.81%). Data preprocessing variant 0.0 had the highest accuracy (75.16%), followed by variant 4.0 (75%). The Random Forest model showed the highest accuracy but tended to overfit, while SVM was more balanced in terms of precision and recall. SVM also processed faster than Random Forest, with the STFT feature set having the fastest processing time.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	muscle fatigue, surface EMG, early detection, STFT, CWT, SVM, RF, edge, kelelahan otot, surface EMG, deteksi dini, STFT, CWT, SVM, RF, edge
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QP Physiology T Technology > T Technology (General) T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5102.9 Signal processing.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Computer Engineering > 90243-(S1) Undergraduate Thesis
Depositing User:	Muhammad Fauzan Nur Insan Pribadi
Date Deposited:	05 Aug 2024 05:06
Last Modified:	05 Aug 2024 05:06
URI:	http://repository.its.ac.id/id/eprint/111934

Actions (login required)

View Item