Detecção da ocorrência de detonação em motores de combustão interna de ignição por centelha baseada em redes neurais artificiais

Abstract

In this work, knock detection techniques were evaluated using the cylinder pressure signal and the mechanical vibration signal in a spark ignition internal combustion engine. Tests were performed to obtain a dataset with operating points in conditions with and without knock. Two signals were acquired on the engine: the cylinder pressure signal and the signal from the mechanical vibration sensor attached to the engine block. To the cylinder pressure signal, it was applied the signal energy pressure oscillations (SEPO) technique for knock detection standardization. To the mechanical vibration signal, it was applied the adapted classical maximum amplitude of vibration (MAV) and signal energy of vibration (SEV) techniques. An artificial neural network (ANN) technique was developed for the improvement of knock detection by employment of the mechanical vibration sensor. A higher agreement rate was reached by the ANN technique over the best case scenario obtained by the classical techniques, where it was possible to obtain a lower percentage of detection errors by the ANN technique in comparison to MAV and SEV. Specifically, the ANN technique obtained a total of 204% reduction in the detection errors over MAV, whereas it reached a 57% reduction over SEV. Therefore, the ANN approach developed in this study is promising as an effective tool for reaching higher agreement rates of detection of knock, thus, increasing the efficiency of spark ignition internal combustion engines.