Detection of Knocking Combustion Using the Continuous Wavelet Transformation and a Convolutional Neural Network

The phenomenon of knock is an abnormal combustion occurring in spark-ignition (SI) engines and forms a barrier that prevents increase thermal efficiency while simultaneously reducing CO2 emissions. Since knocking highly stochastic, cyclic analysis in-cylinder pressure necessary. In this study we propose approach for efficient robust detection identification three different internal engines. proposed methodology includes signal processing technique, called continuous wavelet transformation (CWT), which provides simultaneous the traces time frequency domains with coefficients. These coefficients serve as input convolutional neural network (CNN) extracts distinctive features performs image recognition task order to distinguish between non-knock knock. results revealed following: (i) CWT delivered stable effective feature space represents unique time-frequency pattern each individual cycle; (ii) was superior state-of-the-art threshold value exceeded (TVE) method maximum amplitude oscillation (MAPO) criterion improving overall accuracy by 6.15 percentage points (up 92.62%); (iii) + CNN does not require calibrating values or operating conditions long enough diverse data used train network.