Spark advance self-optimization with knock probability threshold for lean-burn operation mode of SI engine

Chance-Constrained Optimization for Torque Tracking Control with Improving Fuel Economy in Spark-Ignition Engines

“…Clearly, the combustion process will be retarded as the SA of the cycle is decreased, namely, CA50, the crank angle after top dead center (ATDC) when the burnt percent reach 50% is larger. However, due to the uncertainty of the combustion phenomenon, the effect of SA on the combustion process is not deterministic even at a fixed SA value [15], also as shown in Fig. 2, although the SA is the same, the curves are different due to the uncertainty in combustion process.…”

“…where S A k max is calibrated as the boundary where there is no knock event and much more retarded than S A max in (15). The results are shown in Fig.…”

“…6 under which the generated torque is the same. Besides, not all the points on a line with the same torque is feasible since CA50 should be constrained in the interval [6], [15]. The control design problem considered in this paper is to find a feedback torque tracking control law with high combustion quality in the sense of rejecting the stochastic variation of combustion phase by constraint of probability of CA50 hanging out a prespecified range.…”

A survey on online learning and optimization for spark advance control of SI engines

“…These factors result in the varying boundary of physical constraints. The on-board implementation of online learning and optimization algorithms would allow a real-time handling of the varying stochastic boundary for every operation condition [11].…”

“…Three classes of ES are successively reviewed: stochastic approximation-based ES [7,[12][13][14][15][16][17][18][19][20], sinusoid-based ES [21][22][23][24][25][26][27][28][29][30][31][32][33][34][35], and natural perturbation-based ES [36][37][38][39]. This article then focuses on stochastic threshold control algorithms for iterative solution of probability control problems in knock limit control applications [2,10,11,[40][41][42][43][44][45]. Two state-of-the-art stochastic limit control algorithms are reviewed: likelihood-based algorithm and statistical learning-based algorithm.…”