Riassunto analitico
This thesis work was carried out thanks to the collaboration between Maserati S.p.A. and Alma Automotive Srl, more precisely in the Powertrain department, Base Engine group, which deals with calibration and experimentation of gasoline engines. The purpose of this activity is the development of a tool for the offline calibration of ECU (Electronic Control Unit) models. The idea was to provide an alternative method to the standard calibration process: nowadays, greater part of the calibration activities are performed at the test bench and on the road. The method we are discussing in this work is oriented to save time and money by the automation of the calibration process, performing it offline and using advanced algorithms such as neural networks. In particular, the focus was on the Gasoline Particulate Filter (GPF) model and on the Raw Soot Emission model. The latter model was implemented in Simulink. On the other hand, the development of not fully accessible GPF functions (Filtration Efficiency and Soot Regeneration) were carried out using the Deep Learning Toolbox of MATLAB (Neural Networks), which represent a solution for black box modelling. Once the models have been developed, they were tested according to different operating conditions. The Raw Soot Emission model showed high potentiality for Simulink models obtaining more than satisfying results, especially in critical operating conditions such as cold starts. The neural network models exhibit good results considering a single set of calibrations, while a path for future improvements of the activity was defined in order to consider several sets of calibrations.
|
