Statistical engine knock modelling and adaptive control

Abstract: A new statistical concept of the knock control of a spark ignition automotive engine is proposed. The control aim is associated with the statistical hypothesis test which compares the threshold value with the average value of the maximum amplitude of the knock sensor signal at a certain frequency. Achievement of the control aim implies the desired separation between the average value of the maximum amplitude and the target value and hence the desired probability of knock occurrence. This new control concept allows the control algorithm parameters to be connected to the probability of knock occurrence and customer-related data. The regulation error is defined as the difference between the actual and the desired values of the statistic utilized. A control algorithm which is used for minimization of the regulation error realizes a simple count-up–count-down logic. A new adaptation algorithm for the knock detection threshold is developed. The confidence interval method is used as the basis for adaptation. A knock detection threshold is presented using a confidence interval with a certain significance level. The adaptation is performed for an aged engine so that the significance level is the same for the new and the aged engines despite the fact that the detection threshold values are different. This, in turn, guarantees the same knock detection performance for new and aged engines. A simple statistical model which includes generation of the amplitude signals, threshold value determination, and a knock sound model is developed for evaluation of the control concept. The statistical knock audibility concept is associated with the outlier detection method and is used in this paper for knock audibility judgement. A Volvo six-cylinder prototype engine equipped with cylinder pressure and block vibration sensors was used in the experiments. An external microphone was used for the knock sound measurements