Nitrogen oxides emissions are an important aspect of engine design and calibration due to increasingly strict legislation. As a consequence, accurate modeling of nitrogen oxides emissions from Diesel engines could play a crucial role during the design and development phases of vehicle powertrain systems. A key step in future engine calibration will be the need to capture the nonlinear behavior of the engine with respect to nitrogen oxides emissions within a rapid-calculating mathematical model. These models will then be used in optimization routines or on-board control features.
In this paper, an artificial neural network structure incorporating a number of engine variables as inputs including torque, speed, oil temperature and variables related to fuel injection is developed as a method of predicting the production of nitrogen oxides based on measured test data. A multi-layer perceptron model is identified and validated using data from dynamometry tests.
The model predicts exhaust nitrogen oxide concentrations under different engine conditions with satisfactory accuracy. The developed neural network model has potential applications in real-time control aimed at reducing nitrogen oxides emission levels.