Strategi Pengaturan Komposisi Hydrogen dan Diesel Fuel pada Hybrid Hydrogen-Diesel Engine Menggunakan Machine Learning

Guntoro, Nicholas Satria Putra (2025) Strategi Pengaturan Komposisi Hydrogen dan Diesel Fuel pada Hybrid Hydrogen-Diesel Engine Menggunakan Machine Learning. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5018211038-Undergraduate_Thesis.pdf Download (6MB)

Abstract

Penelitian ini mengembangkan pendekatan prediktif berbasis machine learning untuk mengatur komposisi bahan bakar hidrogen dan diesel dalam mesin hybrid. Tujuannya adalah meningkatkan efisiensi pembakaran dan menurunkan emisi gas buang. Empat model machine learning yang digunakan meliputi XGBoost Regressor, Random Forest Regressor (RFR), Multi-Layer Perceptron (MLP), dan Deep Neural Network (DNN). Setiap model dievaluasi menggunakan metrik R², Mean Absolute Error (MAE), dan validasi hold-out. Hasil menunjukkan bahwa XGBoost memberikan performa terbaik dengan nilai R² sebesar 0,9538, MAE 2,77, dan waktu komputasi hanya 0,39 detik. Sementara itu, DNN menghasilkan R² sebesar 0,8077 dan MAE 9,16 setelah 50 kali pelatihan, serta membutuhkan waktu komputasi 3671 detik. Model terbaik (XGBoost) kemudian digunakan dalam simulasi interaktif untuk memprediksi performa mesin secara real-time berdasarkan input throttle. Visualisasi output seperti rasio hidrogen, emisi CO₂ dan NOₓ, serta tingkat asap memberikan pemahaman mendalam terhadap dinamika sistem pembakaran. Penelitian ini menunjukkan bahwa machine learning mampu menjadi solusi efektif dalam optimasi sistem bahan bakar hybrid dan berpotensi diterapkan dalam sistem kendali mesin kapal masa depan.
========================================================================================================================================
This study develops a predictive machine learning approach for optimizing hydrogen and diesel fuel composition in hybrid engines. The aim is to enhance combustion efficiency while reducing exhaust emissions. Four machine learning models were employed: XGBoost Regressor, Random Forest Regressor (RFR), Multi-Layer Perceptron (MLP), and Deep Neural Network (DNN). The models were evaluated using R², Mean Absolute Error (MAE), and hold-out validation. Results indicate that XGBoost achieved the best performance with an R² score of 0.9538, MAE of 2.77, and a training time of only 0.39 seconds. In comparison, DNN reached an R² of 0.8077, MAE of 9.16, required 50 training iterations, and a total computation time of 3671 seconds. The best-performing model (XGBoost) was integrated into a real-time interactive simulation that predicts engine performance based on throttle input. Visualization of outputs such as hydrogen ratio, CO₂ and NOₓ emissions, and smoke level enhances understanding of combustion dynamics. This study demonstrates the effectiveness of machine learning in hybrid fuel system optimization and its potential in future intelligent marine engine control.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Mesin Hybrid, Bahan Bakar Hidrogen-Diesel, Machine Learning, Efisiensi Pembakaran, Reduksi Emisi, Model Prediktif, XGBoost, Random Forest Regressor, MLP, Simulasi,Hybrid Engine, Diesel-Hydrogen Fuel, Machine Learning, Fuel Efficiency, Emission Reduction, Predictive Model, XGBoost, Random Forest Regressor, MLP, Simulation
Subjects:	T Technology > TJ Mechanical engineering and machinery > TJ164 Power plants--Design and construction T Technology > TJ Mechanical engineering and machinery > TJ213 Automatic control. T Technology > TJ Mechanical engineering and machinery > TJ217 Adaptive control systems V Naval Science > VM Naval architecture. Shipbuilding. Marine engineering > VM731 Marine Engines
Divisions:	Faculty of Marine Technology (MARTECH) > Naval Architecture and Shipbuilding Engineering > 36201-(S1) Undergraduate Thesis
Depositing User:	Nicholas Satria Putra Guntoro
Date Deposited:	07 Aug 2025 05:28
Last Modified:	07 Aug 2025 05:28
URI:	http://repository.its.ac.id/id/eprint/127107

Actions (login required)

View Item