Tuning and experimental evaluation of a likelihood-based engine knock controller

Abstract

Abstract-A new likelihood-based stochastic knock controller, that achieves a significantly improved regulatory response relative to conventional strategies, while also maintaining a rapid transient response is presented. Up until now it has only been evaluated using simulations and the main contribution here is the implementation and validation of the knock controller on a five cylinder engine with variable compression ratio. Furthermore, an extension of the fast response strategy and a re-tuning of the controller is shown to improve performance. The controller is validated with respect to its robustness to changes in engine operating condition as well as compression ratio. The likelihood-based controller is demonstrated in engine tests and compared to a conventional controller and it is shown that it is able to operate closer to the knock limit with less variations in control action without increasing the risk of engine damage.

Stochastic simulation and analysis of a classical knock controller

“…[13][14][15][16] In particular, a recent emphasis on knock as a stochastic cyclic process has proved fruitful in the development of new likelihood or cumulative-summation-based knock control strategies and their optimization. [17][18][19][20][21] Despite these developments, however, there currently exist no tools that embrace the stochastic nature of knock when quantifying and comparing the performance of knock controllers. Such quantification is not trivial since repeating an experiment, whether on an engine test bed, or in simulation, is likely to yield a different response according to the random arrival of knock events in that particular instance.…”

“…13–16 In particular, a recent emphasis on knock as a stochastic cyclic process has proved fruitful in the development of new likelihood or cumulative-summation-based knock control strategies and their optimization. 17–21…”

Experimental Validation of a Likelihood-Based Stochastic Knock Controller

“…The response to knocking events, meanwhile, is minimally affected. In the second case, this strategy is extended by resetting the cycle count to a value which increases with the number of 'repeat advance' adjustments that have been made without the occurrence of a knock event, see [23] for further details. The reset cycle count is increased in three stages, after one, two and three spark advances with zero knock events, so that the rate of spark advance accelerates until the first knock event occurs.…”

“…The main contribution of this paper is therefore the experimental implementation and validation of the new control strategy. Preliminary results from this work were presented in [23], but this paper provides more detail. Furthermore, in this paper the robustness of the method is tested by implementing the strategy on two different engines, at two different institutions, and using two different metrics for quantifying knock intensity.…”

“…This strategy has a proven record for engine protection against engine damage but it operates at a mean spark advance which is significantly retarded relative to the desired BorderLine (BL) knock limit target, and therefore suffers some efficiency penalty. In a series of recent papers [20][21][22][23][24][25][26], the authors have advocated a new approach to knock control based on the fundamental binomial properties of knock events. Unlike the traditional controller which reacts deterministically to each knock event, the new controllers only adjust the spark when the statistical likelihood ratio of the observed knock events falls below some user-specified threshold.…”