This paper deals with physical modelling of a turbocharger with variable turbine geometry (VTG) and its real-time simulation based on dynamic artificial neural networks (ANN).
Thermodynamic und fluiddynamic equations, describing the basic functionality and relations between pressure, mass flow and temperature at the inlet and outlet ports of compressor and turbine, build up a multiple input multiple output model (MIMO).
A special kind of ANN, namely the LoLiMoT algorithm, is used for real-time simulation. Training the network using measurement and simulated data, the dynamic behaviour can be simulated with less computational effort than the physical model.
The neural network may be used in engine control systems as observer for non measurable signals, like rotor speed or turbine and compressor torque, figure 1.