Modeling of Unburned Hydrocarbon Emission in a Di Diesel Engine Using Neural Networks

The reduction in the toxic pollutant emissions in diesel engine exhaust gas is one of the main tasks of designers. Toxic substances emitted in diesel exhaust are, among other things, hydrocarbons. Reducing their emissions can be achieved by affecting the exhaust gases or reducing their formation in the combustion chamber. One method is to change the control parameters of the fuel injection process. In the present study, the direct injection diesel engine with a displacement of 1850 ccm was tested. The diesel engine was equipped with a prototype common rail injection system, allowing for the injection of a fuel quantity divided into three parts during one engine working cycle. Each part can be injected when the injector opens at certain injection advance angles. This results in six different control parameters. Such a number of parameters means a large number of combinations. Therefore, in order to cover the entire operating range of the engine, a number of measurements should be carried out. The application of the PS/DS-P: λ test plan has significantly reduced the number of necessary tests. Based on the test plan, learning data for the neural network was obtained. The next step was to develop and educate others on the structure of the neural network. The neural network allows for generation of output data for the input data of the network outside the training set. The model verification carried out on a test engine had satisfactory results. The relative uncertainty did not exceed 8%, while the character of changes in the network output data was also kept when changing the input parameters. The developed model can be used in simulation tests of a diesel engine. In order to facilitate the identification of the obtained data, the results of simulation tests were presented on three-dimensional graphs.