Development and Software-in-the-Loop Validation of an Artificial Neural Network-Based Engine Simulator

Due to the ever increasingly stringent emission regulations for passenger vehicles, the efficiency and performance increase of Spark Ignition (SI) engines have been under the focus of the engine manufacturers. The quest for efficiency and performance increase has led to the development of increasingly complex powertrains and control strategies. The development process requires novel methods that feature a smooth transition between the real and the virtual prototypes. Furthermore, to reduce the development time and cost, developing an engine simulator with a low computational effort and good accuracy, which predicts the engine behavior on the entire operating range, plays a crucial role. This work proposes an Artificial Intelligence-based engine simulator for a Spark Ignition engine. The simulator relies on Neural Networks for the calculation of the main combustion metrics. In the first part of this paper, the data acquired at the engine test cell are analyzed. A shallow neural network model is set up in Matlab for modelling the combustion phase and the knock intensity. The training of the network is performed with a Design of Experiment (DoE) approach, where different numbers of neurons are tested using training algorithms and activation functions. The output of the simulation is then compared to the experimental values. The results are evaluated using the Root Mean Square Error and R-square indexes, and the combination of the number of neurons, training algorithm, and activation function, which yield the lowest error and the highest R-square, are selected. In the second part, the ANN models are then coupled to analytical sub-models previously developed and individually validated, to build a complete engine simulator, and they are implemented in Simulink. The performance of this simulator is then evaluated by comparing simulated results and experimental data, both under steady-state and transient conditions.