The paper deals with the simulation of the wall wetting dynamics in SI engines, making use of Recurrent Neural Networks (RNN). RNN are derived from the Multi Layer Feedforward Neural Networks, largely adopted for static mapping, by considering feedback connections between output and input layers. A Multi Input-Single Output structure has been adopted, assuming injected fuel, manifold pressure and engine speed as external input variables; the Air-Fuel Ratio at the exhaust gas oxygen sensor location has been considered as system output. The RNN has been trained (i.e. identified) and tested vs. a set of transient data measured on a commercial 4 cylinders SI engine at the test bench. The results show a good level of accuracy confirming the suitability of RNN for both HIL simulation or off-line identification of classical Mean Value Models with a drastic reduction of the calibration effort.
Experimental Validation of a Recurrent Neural Network for Air-Fuel Ratio Dynamic Simulation in S.I. I.C. Engines
ARSIE, Ivan;PIANESE, Cesare;SORRENTINO, MARCO
2004-01-01
Abstract
The paper deals with the simulation of the wall wetting dynamics in SI engines, making use of Recurrent Neural Networks (RNN). RNN are derived from the Multi Layer Feedforward Neural Networks, largely adopted for static mapping, by considering feedback connections between output and input layers. A Multi Input-Single Output structure has been adopted, assuming injected fuel, manifold pressure and engine speed as external input variables; the Air-Fuel Ratio at the exhaust gas oxygen sensor location has been considered as system output. The RNN has been trained (i.e. identified) and tested vs. a set of transient data measured on a commercial 4 cylinders SI engine at the test bench. The results show a good level of accuracy confirming the suitability of RNN for both HIL simulation or off-line identification of classical Mean Value Models with a drastic reduction of the calibration effort.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.