Volumetric efficiency estimation based on neural networks to reduce the experimental effort in engine base calibration

Over the last decades, internal combustion engines have undergone a continuous evolution to achieve better performance, lower pollutant emissions and reduced fuel consumption. This evolution involved changes in the engine architecture needed to perform advanced management strategies. Therefore, Variable Valve Actuation, Exhaust Gas Recirculation, Gasoline Direct Injection, turbocharging and powertrain hybridization have widely equipped modern internal combustion engines. However, the effective management of a such complex system is due to the contemporaneous development of the on-board Engine electronic Control Unit. In fact, the additional degrees of freedom available for the engine regulation highly increased the complexity of engine control and management, resulting in a very expensive and long calibration process. For this reason, this study proposes an effective methodology based on the use of Neural Networks to overcome some critical issues concerning the calibration of engine control parameters. NN are adopted to provide a detailed engine data sheet starting from a reduced number of experimental data. To verify the potential of the proposed methodology, this detailed data set is subsequently used as input to a specific Computer Aided Calibration algorithm developed by the authors and the achievable calibration performance are evaluated. In particular, the calibration performance was assessed with reference to a specific ECU function in this paper. The research clearly demonstrates the effectiveness of the proposed approach since the calibration performance falls within acceptable limits even after a 60% cut of the experimental data usually acquired for calibration purposes, highlighting how the use of neural networks can allow a significant reduction of the experimental effort along with its related times and costs.