Computing the closed-loop characteristics of a generalized multi-threshold knock controller

Abstract

Most knock controllers respond to knock events which are defined according to some threshold knock intensity. Multi-threshold knock events offer more informative feedback since they encode not just the occurrence of knock events but also some measure of their intensity. While this has the potential for improved control, it is hard to assess the extent to which any benefits are truly realized because (in common with all knock control systems) the results of any single experiment or simulation depends on the random arrival of knock events in that instance. In this article, methods are developed instead to compute the statistical properties of the closed-loop response of a general multi-threshold knock controller, thereby providing a much more complete and rigorous characterization of its performance than has previously been possible. The method is applied to single- and dual-threshold knock controllers and used to provide a rigorous comparison of the transient and steady-state performance of these different control laws. The method can also be used as a calibration aid to assess the effects of different controller gains in reliable, repeatable fashion.

A critical analysis of classical severity-based knock control strategies

“…The knock severity control strategy defined by (4), (5), is relatively straightforward to implement, but it is harder to analyze because control actions now depend on knock intensity as well as the knocking/non-knocking status of the prior cycle. Discretizing knock intensities according to (1), however, enables (4), (5) to be cast in the form of a multi-threshold controller, 19…”

A fuzzy logic map-based knock control for spark ignition engines

“…Recently, a new dual-threshold knock controller was presented in [17] where the binary knocking and no-knocking classification used by most knock-eventbased controllers was extended to a ternary knocking/notknocking/indeterminate classification using two knock thresholds, this reduced the chance of misclassifying knocking cycles as non-knocking. Reducing the misclassification of knocking events allows to increase the controller gain, which leads to faster response [18]. A further method is presented in [19], where a beta distribution based stochastic knock control is proposed, knock probability estimation is performed by a Bayes rule and beta distribution and likelihood ratio test is used for control decision.…”

On-board knock probability map learning–based spark advance control for combustion engines

“…The likelihood-based stochastic knock controller (LHC) and its variants are representative knock probability controllers that have been developed in recent years. 11–15 The LHC operates SA at the borderline with acceptable knock probability in a closed-loop control scheme. The knock probability is estimated over a period of cycles by maximum likelihood estimation (MLE) according to the online measured binary knock signal, supposing that knock event obeys Bernoulli’s distribution.…”