Development of a Machine Learning-Driven Predictive Model for Soot/NO<sub>X</sub> Emissions in Reduced Biodiesel Chemical Mechanisms for Internal Combustion Engines

Song, Yunseok; Shin, Cheolsu; Suh, Hyun Kyu

doi:2026.27.4.1759

International Journal of Automotive Technology > Volume 27(4); 2026 > Article

Electric, Fuel Cell, and Hybrid Vehicle, Engine and Emissions, Fuels and Lubricants, Heat Transfer, Fluid and Thermal Engineering

International Journal of Automotive Technology 2026;27(4): 1759-1768.
doi: https://doi.org/10.1007/s12239-025-00389-4

Development of a Machine Learning-Driven Predictive Model for Soot/NO_X Emissions in Reduced Biodiesel Chemical Mechanisms for Internal Combustion Engines

Yunseok Song¹, Cheolsu Shin², Hyun Kyu Suh³

¹Korea Automotive Technology Institute, 74 Yongjung-ri, Pungse-myeon, Cheonan, Korea
²Graduate School of Mechanical Engineering, Kongju National University, 1223-24, Cheonan-daero, Seobuk-gu, Cheonan, 31080, Korea
³Division of Mechanical and Automotive Engineering, Kongju National University, 1223- 24, Cheonan-daero, Seobuk-gu, Cheonan, 31080, Korea

Corresponding Author. Hyun Kyu Suh , Email. hksuh@kongju.ac.kr

Received: August 18, 2025; Revised: August 18, 2025 Accepted: September 24, 2025. Published online: December 10, 2025.

ABSTRACT

In this study, a biodiesel reduced chemical reaction mechanism was developed, optimized for 3D CFD engine analysis. The mechanism was condensed to focus on the low-temperature region, where fuel oxidation begins. By optimizing the reaction rates of key reactions identified through sensitivity and reaction path analysis, errors in ignition delay, in-cylinder pressure, heat release rate, and exhaust emissions (soot and NO_X) were minimized. Furthermore, a long short-term memory (LSTM) machine learning model was also developed and evaluated for predicting 0D results. The mechanism reduction resulted in an 89% decrease in chemical species and reactions and a 95% reduction in analysis time, though the merging of solid-phase soot increased the time by 70%. Optimizing the reaction rates reduced the 0D ignition delay error and the 3D CFD errors for pressure (from 15.01 to 2.418%), heat release (from 35.86 to 9.664%), soot (3.69%), and NO_X (3.53%). However, focusing on the low-temperature region increased errors in the high-temperature region, showing the challenge of achieving simultaneous improvement. The machine learning model achieved a 97% prediction accuracy for temperature, pressure, NO, NO2, and CO, but had 13% error for fuel chemical species and VOCs, indicating difficulties in predicting complex species.

Key Words: Biodiesel · Chemical reaction mechanism · Mechanism reduction · Low temperature region · Ignition delay · Exhaust emissions · Machine learning