Abstract
Engines/vehicle systems are becoming increasing complex partly due to the incorporation of emission abatement components and associated control and reagent management strategies that are technologically evolving 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 and control concepts.
Two methods are described in which detailed kinetics models for emissions abatement catalysts and detailed engine models are transformed into a fast running model for control design, calibration or real-time ECU validations. The first methodology is based on a structured, semi-automatic scheme for reducing each high-fidelity model into a fast running neural network (NN) model by training it using simulation results from detailed models. The NN model inputs are identical to those used in the detailed model. However, there are some additional inputs that need to be calculated or modeled. In particular, NH3/NOx usage ratio in an SCR is included as a dependant variable of input stream state (concentration, flow rate and temperature) and the state of the catalyst (coverage and temperature). In this regard, the catalyst state is calculated through two physical models. The second methodology is developed such that each subsystem runs at its optimum time step size using the appropriate solver while boundary conditions (species flow rate, pressure and temperature) are dynamically shared between subsystem models. A special connection (circuit splitter) is implemented for this purpose. Different configurations and levels of details are presented to demonstrate the flexibility of the methodology. The comparison of results shows that both methods are able to conserve the accuracy as well as achieve computational efficiency at appropriate system design stages. This makes advanced engine control design, calibration and ECU validation (involving coupled engine with aftertreatment) feasible and computationally efficient while preserving the flexibility requirement for concept evaluations.
Keywords: system model, neural network, NN, HiL, SiL, aftertreatment, catalyst, DOC, SCR