A Model-Based Computer Aided Calibration Methodology Enhancing Accuracy, Time and Experimental Effort Savings Through Regression Techniques and Neural Networks

In the last few years, the automotive industry had to face three main challenges: compliance with more severe pollutant emission limits, better engine performance in terms of torque and drivability and simultaneous demand for a significant reduction in fuel consumption. These conflicting goals have driven the evolution of automotive engines. In particular, the achievement of these mandatory aims, together with the increasingly stringent requirements for carbon dioxide reduction, led to the development of highly complex engine architectures needed to perform advanced operating strategies. Therefore, Variable Valve Actuation (VVA), Exhaust Gas Recirculation (EGR), Gasoline Direct Injection (GDI), turbocharging, powertrain hybridization and other solutions have gradually and widely been introduced into modern internal combustion engines, enhancing the possibilities of achieving the required goals. However, none of the improvements would have been possible without the contextual development of electronics. In fact, that solutions have highly increased the complexity of engine control and management because of the degrees of freedom available for the engine regulation, thus resulting in a long calibration time. In particular, base calibration is the most onerous phase of the engine control, both in terms of experimental and computational effort and costs. This paper addresses some critical issues concerning the calibration of control parameters through the use of a specific Model-Based Computer Aided Calibration algorithm developed by the authors to automate the calibration process and minimize calibration errors. The proposed methodology is also based on the use of neural networks (NN). In particular, starting from a reduced number of experimental data, NN provide a detailed engine data sheets used as input to the actual calibration process itself. The proposed algorithm provides optimal portability and reduced calibration time. The research also highlights how the developed methodology could be useful to identify possible enhancements for specific ECU engine models that can improve the accuracy of the calibration process by using more detailed physically based functions. The results of the proposed research clearly highlight how, in engine control, more accurate physical modeling may lead to promising results and better performance, ultimately enhancing the accuracy, time, experimental effort and cost savings of the calibration process.