Artificial Neural Network Modelling of Performance and Emissions in Turbulent Biodiesel Combustion within a Compression Ignition Engine

Abstract

Accurate prediction of engine performance and emissions is essential for optimizing biodiesel-fueled compression ignition engines under varying operating conditions. This study develops an artificial neural network (ANN) model to estimate in-cylinder temperature, specific fuel consumption (SFC), brake efficiency, hydrocarbon (HC), carbon monoxide (CO), nitrogen oxides (NOₓ), and particulate matter (PM) based on key engine operating parameters. Experimental data covering a wide range of engine loads and speeds were used for training, validation, and testing of multiple multilayer perceptron (MLP) networks. The ANN demonstrated strong predictive capability, with correlation coefficients ranging from R = 0.755 to 0.997 and low root mean square errors, such as 21.33 °C for in-cylinder temperature and 8.49 ppm for CO. Excellent agreement was observed for SFC, brake efficiency, CO, and NOₓ, while moderate deviations in HC and PM were attributed to their inherently stochastic formation processes. These results confirm that the ANN provides a reliable, computationally efficient tool for modeling engine performance and emissions. The model can support engine optimization studies, and future work will focus on incorporating fuel physicochemical properties and hybrid modeling approaches to further enhance predictive accuracy.