Artificial Neural Network Based Emissions Modeling for Spark Advance Optimization in an Ammonia Fueled Engine

The reduction of carbon dioxide emissions is a key objective in the transition toward more sustainable energy and transportation systems. Among the various strategies under investigation, the use of alternative fuels represents a viable pathway to reduce the carbon footprint of internal combustion engines. In this context, ammonia has attracted increasing in- terest as a carbon neutral fuel, as it contains no carbon atoms and therefore enables combustion without direct CO2 emissions. Despite this advantage, the adoption of ammonia introduces sig- nificant challenges. Ammonia is a toxic compound and its release into the environment must be strictly controlled. In particular, unburned ammonia NH3, NOx and N2O emissions represent critical issues, requiring accurate prediction and effective con- trol of the combustion process to ensure both environmental compliance and stable engine operation. This work proposes a data driven approach for emissions estimation and control in an ammonia fueled internal combustion engine. In the first stage, a predictive model based on artificial neural networks is developed to estimate the main exhaust emissions as a function of key engine operating parameters. The model is trained and validated on experimental datasets, enabling the capture of the highly nonlinear relationships between control inputs and emission outputs. In the second stage, the trained neural network model is integrated into an optimization framework aimed at improving engine control strategies. In particular, control variables such as spark advance are tuned to minimize pollutant emissions.