Artificial Neural Network-Based Performance and Emission Prediction for Hydrogen-Combustion Multicylinder Engines Federal University of Santa Maria

Format:

Book

Conference/Event

Author/Creator:

Pasa, Bruno Roberto, author.

Contributor:

Fagundez, Jean Lucca Souza

Lanzanova, Thompson Diórdinis Metzka

Martins, Mario Eduardo Santos

Salau, Nina Paula Gonçalves

Silveira, Juliano Pereira

Conference Name:

SAE Brasil 2025 Congress (2025-10-07 : Sao Paolo, Brazil)

Language:

English

Physical Description:

1 online resource cm

Place of Publication:

Warrendale, PA SAE International 2025

Summary:

In response to the pressing need to reduce greenhouse gas emissions from the transportation sector, hydrogen-fueled internal combustion engines (H2ICEs) have emerged as a promising alternative to conventional fossil-fueled powertrains. However, optimizing H2ICEs presents challenges in balancing performance with emissions, particularly in nitrogen oxide (NOx) formation This study proposes a data-driven methodology using an artificial neural network (ANN) to predict key emission and performance metrics: NOx emissions, brake mean effective pressure (BMEP), brake specific fuel consumption (BSFC), brake power, and brake thermal efficiency, based solely on engine operational parameters. Experimental data were collected from a three-cylinder Ford EcoBoost engine under varying conditions of intake pressure, spark timing, air-fuel ratio, engine speed, and valve timing. Feature selection was performed using the Spearman correlation coefficient, identifying engine speed, start of injection angle (SOI), air-fuel ratio (λ), and intake pressure as the most important input variables. Bayesian optimization was employed to tune the ANN's hyperparameters, resulting in a network architecture with a single hidden layer consisting of 10 neurons using the tanh activation function, optimized with the Adam optimizer at a learning rate of 0.01. The final ANN model exhibited satisfactory predictive performance, achieving correlation coefficients greater than 0.97 for most outputs and exceeding 0.95 across all predicted variables. These results demonstrate that the proposed ANN effectively captures the nonlinear behavior of hydrogen-fueled engines and offers a valuable tool for reducing the experimental burden in engine calibration and development, thereby supporting the advancement of hydrogen-powered mobility solutions

Notes:

Vendor supplied data

Publisher Number:

2025-36-0237

Access Restriction:

Restricted for use by site license