Greybox modeling of the diesel combustion by use of the scalar dissipation rate

This paper deals with greybox modeling of a diesel combustion process for the purpose of model-based combustion control. For this, a real-time capable model with a high prediction quality and a high interpolation and extrapolation capability is required. An existing neural network model of the process will be introduced. The model inputs are the quantity of injected fuel, start of injection and the intake manifold fraction of recirculated exhaust gas. The variables to be predicted are the position of the combustion phasing, the indicated mean effective pressure and the maximum cylinder pressure gradient, which correlates with the combustion noise of the engine. To enhance the existing model, an analytic description of the scalar dissipation rate is used as a new input into the neural network. The aim here is to decrease the model error, improve the model robustness and reduce the complexity of the neural network. The scalar dissipation rate describes the diffusivity in mixture fraction space and represents a combustion characteristic quantity. For a couple of engine operating points, the scalar dissipation rate is calculated by 3D CFD simulations. Afterwards, the calculation results are utilized to find an analytic description. To use this new model information a restructuring of the neural network is necessary. The modeling results will be shown and compared to the existing blackbox model.