Author ORCID Identifier

https://orcid.org/0009-0000-2882-6945

Semester

Spring

Date of Graduation

2026

Document Type

Thesis

Degree Type

MS

College

Statler College of Engineering and Mineral Resources

Department

Mechanical and Aerospace Engineering

Committee Chair

V’yacheslav Akkerman

Committee Co-Chair

Almoutazbellah Kutkut

Committee Member

Hailin Li

Abstract

The increasing stringency of emissions regulations for heavy-duty (HD) diesel engines necessitates the development of accurate, efficient, and general predictive models for engine-out emissions under steady-state operating conditions. Traditional statistical approaches, while computationally efficient, often fail to capture the inherent nonlinear relationships between the engine operating parameters and emissions formation. In contrast, machine learning (ML) techniques offer a promising alternative by learning the complex patterns directly from experimental data. The present thesis summarizes a comprehensive study on the prediction of steady-state, HD diesel engine emissions using ML models, along with a comparative analysis against conventional statistical modeling approaches.

A high-quality experimental dataset comprising multiple steady-state operating conditions was utilized, with key engine inputs including the engine speed, brake torque, intake air flow, and the fuel flow rate. The target outputs include critical emissions parameters such as nitrogen oxides (NOx) concentration, along with additional performance indicators. Two statistical models – the Multiple Linear Regression (MLR) and the Kriging (Gaussian Process Regression) – have been developed as baseline approaches, while three ML models – the Artificial Neural Networks (ANN), the 1D Convolutional Neural Networks (CNN), and the Random Forest (RF) – have been implemented to capture nonlinear dependencies.

The model performance was evaluated using standard metrics such as the coefficient of determination (R2), the root-mean-square error (RMSE) and the mean absolute error (MAE). It is observed that the ML models employed significantly outperform statistical models in predicting emissions, with ANN and CNN exhibiting superior accuracy due to their ability to model complex nonlinear interactions among variables. The RF model also showed strong performance with improved robustness and interpretability. In contrast, MLR exhibited limitations due to its linear assumptions, while Kriging provided improved accuracy but at higher computational cost.

The findings of this study highlight the effectiveness of the ML techniques for steady-state diesel engine emissions prediction and establish a clear performance advantage over traditional statistical methods. This work contributes to the advancement of data-driven modeling approaches for engine development and emissions control, with potential applications in real-time virtual sensing, engine calibration, and regulatory compliance strategies.

Comments

Revised thesis after formatting.

Download

Share

COinS