An analysis of estimation strategies for knock control

Abstract

Since knock behaves as a random process, all knock controllers regulate some statistical property of this process which must be estimated from the data. The choice of the statistical metric used for feedback, its target value, and the method used for its estimation all have a strong impact on the performance of the closed loop system. In this work, expressions for the asymptotic variance of the estimates of different test statistics are derived and used to compare empirical versus parametric estimation methods. The variance of different metrics and the slope of their response curves (e.g. knock probability vs knock percentile intensity feedback) are also used to compute and compare the efficacy of these different measures for knock control purposes. The efficacy is seen to vary as a function of the chosen knock probability target, and lower knock thresholds/higher knock rate targets are recommended. Finally, the estimator dynamics are shown to dominate the dynamics of the closed loop response, and estimates obtained using a sliding window versus an exponentially weighted moving average are compared and contrasted. In each case, the results are illustrated using closed loop simulations of controller response subject to disturbances.