Estimation Method of Knocking Sound and In-cylinder Pressure from Engine Radiation Noise by Deep Learning

Abstract

The deep learning model we developed estimates knocking sound pressure and in-cylinder pressure from engine radiation noise captured by a microphone. This model obtains the time frequency mask and transfer function from many engine radiation noise and in-cylinder pressure pair data. The input to the model is a time-frequency spectrogram of the engine radiation noise calculated by the short time Fourier transform. The time frequency mask extracts the knocking sound from a spectrogram. The transfer function is to convert the extracted knocking sound into an estimated in-cylinder pressure. We propose a knocking detection method that applies this model.