Neural Network Application to Evaluate Thermodynamic Properties of ICE’s Combustion Gases

In this paper, the authors have investigated a new neural network application for the determination of thermodynamic properties for various gases for internal combustion engines applications. The Neural Network has been trained using experimental data available in literature (specific heat at constant pressure, enthalpy, entropy and equilibrium constants for thirteen gases of practical interest inside ICE applications). In the present study a two-layer Elman network feedback from the first-layer output to the first layer input as well as “tansig” neurons in its hidden and out layers has been implemented. After the training, neural network has been tested through a comparison with the NASA equations and JANAF equations, showing the capability to cover with a single model wide range of temperature with an accuracy equal or greater than others mathematical function. Thermodynamic properties of gases have been calculated depending on temperature. In order to evaluate the relative percent error Neural Network thermodynamic results have been compared with experimental data. Neural Networks have been implemented to calculate the thermodynamic properties of several gases: N, O, H, H2, O2, N2, CO, OH, NO, CO2, Ar, N2O and H2O.