Engine Knock Detection Based on Time-Frequency Images and Support Vector Machine

Abstract

Listen

Translate article

Knock is a major cause of pollution, parts damage and metallic noise in an engine. But slight knock can improve the power and economic performance of an engine. Therefore, existence and intensity of the knock are preconditions for the automatic knock control system to work. This paper describes an advanced approach solving the knock detection task. It is based on a time-frequency image generation solution followed by a support vector machine detection step trained in a constructive supervised way. The proposed method is applied to analyze the pressure signals measured from an engine cylinder to detect the knock based on the popular cycle-by-cycle classification. It is shown that this approach can qualify well for knock detection.