Experimental Evaluation of a Bayesian Estimation and Control of Engine Knocking Level

In a context where every percent of consumption improvement counts, focus on engine knock control is still of major concern. It is commonly accepted in the community that knocking phenomenon has a probabilistic nature. Based on this observation, stochastic controllers are the best candidates to manage the engine under knocking conditions. This brief presents a control strategy based on the Bayesian estimation of the distribution quantile of the knock intensity measurements. Despite the quantile estimation randomness, the corrections undertaken by the controller are moderated by the level of confidence in the quantile estimation. Experimental results obtained on a high-efficiency gasoline engine stress the relevance of the approach. First, this strategy ensures a better compromise between the engine consumption and the prevention of knock phenomenon, compared to classical approaches. Second, the results show that the strategy succeeds in managing real-driving conditions.