Engine knock detection: an eigenpressure approach

Abstract

In this work, a knock detection approach based on in-cylinder pressure principal component analysis is proposed. The introduction of a set of basis functions called eigenpressures used to describe the principal components of the pressure traces allows for an easy and effective separation between the typical “bell shape” component of pressure profiles and the knock-induced pressure oscillations, making possible the classification of knocking and not knocking cycles. The proposed approach is compared to a standard knock detection method based on the in-cylinder pressure trace band-pass filtering and to a pure data-driven algorithm. The method shows the best knock classification performances and proves to be advantageous thanks to the low number of easily tunable parameters and their ease of calibration/interpretation.