Estimation of Engine Emissions Based on Physical and Chemical Properties of Biodiesels using Artificial Neural Networks

Research has shown that the emission characteristics of Biodiesels are different compared to petroleum derived diesel. Though overall emissions of most of the Biodiesels are less than that of diesel, it has been found that many bio-esters have higher NO_X emissions. This necessitates the testing of the various blends and selection of the Biodiesel that has acceptable emission characteristics especially with respect to NO_X emissions. Due to the sheer number of variations, it becomes very tedious to manually test for every blend and for every Biodiesel. This paper introduces an elegant method of the above required analysis by establishing an Artificial Neural Network (ANN) that is trained to predict engine emission based on fuel properties.

Emission data is collected by testing a CI engine under different loads for series of blends with diesel of various Biodiesels and is used to train the network. The principal fuel properties-Viscosity, Calorific Value, Flash point, Density, carbon residue and percentage of rated load are used as the input layer of the ANN while the emissions of NO_X, HC, CO, CO₂ and particulates form the output layer. Different network combinations are formed and tested and the optimal design is chosen. ANN codes are written in Matlab software.