Machine Learning Methods for Emissions Prediction in Combustion Engines with Multiple Cylinders

Abstract

The increasing demand of lowering the emissions of the combustion engines has led to the development of more complex engine systems. This paper presents artificial neural network (ANN) based models for estimating nitrogen oxide (NOx) and carbon dioxide (CO2) emissions from in-cylinder pressure of a maritime diesel engine. The architecture of the models is that of Multi-Layer Perceptron (MLP) and Radial Basis Function (RBF) network. The data utilized to train and test the models are obtained from a four-cylinder marine engine. The inputs of the models are chosen as the first principal components of the in-cylinder pressure and engine parameters with highest correlation to aforementioned greenhouse gases. Generalization is performed on the models during the training to avoid overfitting. The estimation result of each model is then compared. Additionally, contribution of each cylinder to the production of emissions is investigated. Results indicate that MLP has a higher accuracy in estimating both NOx and CO2 compared to RBF network. The emission levels of each cylinder for both NOx and CO2 are mostly even due to the nature of the conventional diesel engine.