Modeling diesel particulate emissions with neural networks

Eight different fuels were tested under five steady operating conditions (reproducing the European transient urban/extraurban certification cycle) in a typical European passenger car Diesel engine. The soluble (SOF) and insoluble fractions (ISF) were analyzed using GC and high performance liquid chromatography (HPLC). The influence of the fuel composition parameters (aromatic content, cetane index, gross heat power, nitrogen and sulfur content) on particulate emissions was studied and data were fitted along with operation conditions (torque and engine speed) using neural networks. The mathematical model reproduces experimental data within 87–90% of confidence and allows for the simulation of emissions at steady conditions for any value of parameters in the experimental range. In-house software also allows for the complete estimation of emissions for one single operating mode or for a whole certification cycle including the composition of the ISF (sulfates, nitrates and water) and the total quantity of each fraction. Fuel and air consumption is also estimated from integration; intermediate non-steady conditions are taken into account by considering the acceleration equations.