Modeling and Prediction of Fuel Consumption and Emissions of Direct Injection Diesel Engines using Artificial Neural Network by Applying Control Parameters of Engine Speed, Fuel Mass and Inlet Air Temperature

In the present study, using artificial neural network, modeling and prediction of NOx, soot and fuel consumption in a direct injection diesel engine is done by applying control variables of engine speed, inlet air temperature and fuel mass injected into the combustion chamber. For this purpose, the empirical experiments were carried out to prepare the necessary modeling and correlation between input and output parameters by neural network. The neural network with the Levenberg-Marquardt training algorithm is designed to train the existing relationship between the above parameters, in which the output variables are modeled completely independently. In other words, for any output such as NOx, the number of hidden layer neurons as well as the lattice control parameters would be quite different from the same parameters for soot or fuel consumption. The results show that the designed neural network reaches accuracy 0.97 for 36 neurons in the hidden layer at 3733th epoch to test the data in NOx modeling. Also, modeling of soot with more neurons and accuracy 0.96 is performed in the 2081th epoch. On the other hand, the test accuracy for modeling fuel consumption for 19 neurons in the hidden layer at 3698th epoch was 0.94 which is due to the uneven distribution of the experimental data over a wide range of modeling ranges. The neural network can also be used as an effective method in direct injection diesel engines intelligent control systems to reduce pollutants and fuel consumption due to its fast convergence and hence short response time.