Analisis Performa Metode Levenberg-Marquadt dan Unscented Kalman Filter Pada Identifikasi Parameter Sistem Autonomous Underwater Vehicle

Martsetyo, Dewa Bagus (2025) Analisis Performa Metode Levenberg-Marquadt dan Unscented Kalman Filter Pada Identifikasi Parameter Sistem Autonomous Underwater Vehicle. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5022211219-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Penelitian ini bertujuan untuk menganalisis dan membandingkan performa dua metode estimasi parameter nonlinear, yaitu Nonlinear Least Square Levenberg-Marquardt (NLS-LM) dan Unscented Kalman Filter (UKF), dalam mengidentifikasi parameter hidrodinamika Autonomous Underwater Vehicle (AUV) 5-DOF. Model matematis AUV yang dikembangkan mencakup dinamika rigid-body, gaya hidrodinamika (hidrostatis, redaman, massa tambahan, body lift), dan gaya kontrol dari propeller, rudder, dan stern plane. Metodologi penelitian meliputi simulasi tiga jenis manuver (lurus dengan belokan, angka delapan, dan heliks menurun) dengan variasi kondisi noise untuk mengevaluasi robustness kedua metode.
Hasil penelitian menunjukkan bahwa metode Levenberg-Marquardt memberikan performa superior dengan nilai RMSE berkisar 4.92e-002 hingga 1.00e-001 dalam kondisi noise, dibandingkan UKF yang menghasilkan RMSE hingga 4.48e-001 terutama pada komponen percepatan rotasi. Pada kondisi ideal tanpa noise, LM mencapai estimasi parameter dengan variasi 0% untuk manuver kompleks, sementara UKF menunjukkan variasi parameter ekstrem hingga ribuan persen. Analisis trajectory comparison mengonfirmasi bahwa parameter hasil estimasi LM mampu mereproduksi lintasan AUV dengan akurasi tinggi, sedangkan UKF menghasilkan trajectory yang tidak realistis secara fisik. Kompleksitas manuver terbukti sangat mempengaruhi kualitas estimasi, dimana manuver sederhana menghasilkan parameter yang kurang observable.
Implementasi pada AUV fisik mengalami kendala distribusi massa dan histerisis aktuator. Penelitian ini menyimpulkan bahwa metode Levenberg-Marquardt lebih suitable untuk estimasi parameter hidrodinamika AUV dengan akurasi tinggi, meskipun memerlukan computational cost yang lebih besar dibandingkan UKF.
====================================================================================================================================================================================
This research aims to analyze and compare the performance of two nonlinear parameter estimation methods, namely Nonlinear Least Square Levenberg-Marquardt (NLS-LM) and Unscented Kalman Filter (UKF), in identifying hydrodynamic parameters of a 5-DOF Autonomous Underwater Vehicle (AUV). The developed mathematical AUV model encompasses rigid-body dynamics, hydrodynamic forces (hydrostatic, damping, added mass, body lift), and control forces from propeller, rudder, and stern plane. The research methodology includes simulation of three maneuver types (straight with turn, figure-eight, and descending helix) with varying noise conditions to evaluate the robustness of both methods.
Research results demonstrate that the Levenberg-Marquardt method provides superior performance with RMSE values ranging from 4.92e-002 to 1.00e-001 under noise conditions, compared to UKF which produces RMSE up to 4.48e-001 particularly in rotational acceleration components. Under ideal noise-free conditions, LM achieves parameter estimation with 0% variation for complex maneuvers, while UKF shows extreme parameter variations reaching thousands of percent. Trajectory comparison analysis confirms that LM-estimated parameters can reproduce AUV trajectories with high accuracy, whereas UKF generates physically unrealistic trajectories. Maneuver complexity significantly influences estimation quality, where simple maneuvers result in poorly observable parameters.
Physical AUV implementation encountered challenges with mass distribution and actuator hysteresis. This research concludes that the Levenberg-Marquardt method is more suitable for high-accuracy AUV hydrodynamic parameter estimation, despite requiring higher computational cost compared to UKF.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Autonomous Underwater Vehicle, Sensor Kedalaman, GPS, Nonlinear Leastsquare, Levenberg-Marquadt, Parameter Hidrodinamika, Root Mean Square Error, AUV, Depth Sensor, GPS, Nonlinear Leastsquare, Levenberg-Marquadt, Hydrodynamics Parameter, Root Mean Square Error
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK3070 Automatic control
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20201-(S1) Undergraduate Thesis
Depositing User:	Dewa Bagus Martsetyo
Date Deposited:	30 Jul 2025 07:39
Last Modified:	30 Jul 2025 07:39
URI:	http://repository.its.ac.id/id/eprint/123811

Actions (login required)

View Item