Resumen
This study aims to construct a reduced thermodynamic cycle model with high accuracy and high model execution speed based on artificial neural network training for real-time numerical analysis. This paper proposes a method of constructing a fast average-value model by combining a 1D plant model and exhaust gas recirculation (EGR) control logic. The combustion model of the detailed model uses a direct-injection diesel multi-pulse (DI-pulse) method similar to diesel combustion characteristics. The DI-pulse combustion method divides the volume of the cylinder into three zones, predicting combustion- and emission-related variables, and each combustion step comprises different correction variables. This detailed model is estimated to be within 5% of the reference engine test results. To reduce the analysis time while maintaining the accuracy of engine performance prediction, the cylinder volumetric efficiency and the exhaust gas temperature were predicted using an artificial neural network. Owing to the lack of input variables in the training of artificial neural networks, it was not possible to predict the 0.6–0.7 range for volumetric efficiency and the 1000–1200 K range for exhaust gas temperature. This is because the mean value model changes the fuel injection method from the common rail fuel injection mode to the single injection mode in the model reduction process and changes the in-cylinder combustion according to the injection timing of the fuel amount injected. In addition, the mean value model combined with EGR logic, i.e., the single-input single-output (SISO) coupled mean value model, verifies the accuracy and responsiveness of the EGR control logic model through a step-transient process. By comparing the engine performance results of the SISO coupled mean value model with those of the mean value model, it is observed that the SISO coupled mean value model achieves the desired target EGR rate within 10 s. The EGR rate is predicted to be similar to the response of volumetric efficiency. This process intuitively predicted the main performance parameters of the engine model through artificial neural networks.