A Method for Separating Knocking Sounds from Engine Radiation Noise by Deep Learning

Knocking is the abnormal combustion of a gasoline engine, it generates a metallic noise. Engine knocking can damage the engine, so workers detect knocking by listening to the sound. There is a need to develop a way to automate this kind of work. We developed the deep learning model which separates Knocking sound from engine radiation noise measured by a microphone. This model obtains the time-frequency mask from the paired data of engine emissions and cylinder pressure. The time-frequency mask enables the separation of knocking sound from engine radiation noise. By training various rotation speeds, the proposed model can separate the knocking sound without training target engine speed.