The statistical properties of raw knock signal time histories

Abstract

Traditional processing of knock signals reduces each recorded time history to a single scalar knock intensity metric for each engine firing, thereby losing all information relating to the phasing and evolution of knock over the combustion period. In this work, a statistical analysis of raw knock cylinder pressure and accelerometer signal time histories has been performed showing how the nonstationary stochastic characteristics of these signals evolve as a function of crank angle, and as a function of spark timing. The data are shown to closely approximate a cyclically independent process whose 2-dimensional covariance function captures both the non-stationary deterministic processes taking place within a cycle as well as the stationary random variations from one cycle to the next. A (time-varying) dual-Gaussian model for the distribution of these cyclic variations was fitted to the data, thereby enabling them to be characterized as the sum of knocking and non-knocking populations. The parameters of the model, which describe these populations separately, provide further empirical insight into the knock process.