Engines and vehicle systems are becoming increasing complex partly due to the incorporation of emission abatement components as well as control strategies that are technologically evolving and innovative to keep up with emissions requirements. This makes the testing and verification with actual prototypes prohibitively expensive and time-consuming. Consequently, there is an increasing reliance on Software-In-the-Loop (SIL) and Hardware-In-the-Loop (HIL) simulations for design evaluation of system concepts.
This paper introduces a methodology in which detailed chemical kinetic models of catalytic converters are transformed into fast running models for control design, calibration or real time ECU validation. The proposed methodology is based on the use of a hybrid, structured, semi-automatic scheme for reducing high-fidelity models into fast running models. The resulting hybrid model consists of a set of neural network-based static sub-models that account for the large non-linearity of the system, concatenated with physical sub-models that account for the dynamics and hysteresis that are inherent in the processes being modeled.
A model of DOC-SCR catalyst system was chosen as the surrogate for this methodology. In this regards, the paper will describe procedures involving identification of relevant parameters using experimental data and design of experiment (DOE) optimization. The DOE results were used to train hybrid-NN Model. The comparison of the results show that the methodology conserves accuracy and achieves computational efficiency, thus making advanced engine control design, calibration and ECU validation (involving coupled engine, aftertreatment and vehicle models) in some cases simply feasible, and in other cases more secure, faster and easier.