Combustion Control of Spark-Ignition Engines Based on Map-Learning

Abstract

Combustion phase control is of great importance in spark ignition (SI) engine researches since it affects combustion qualities, such as fuel efficiency, combustion variation, and knocking. Keeping the combustion phase at the optimal reference value considering physical constraints, knock for instance, is challenging due to the engine transient operation, drift of the optimal controllable variable values and varying physical constraints. To address this issue, this research presents an on-board map-learning scheme. Firstly, taking the crank angle of 50% mass burnt (CA50) as the combustion phase indicator, a 3-dimensional mapping from CA50, manifold pressure, and engine speed to thermal efficiency, knock intensity, and spark advance (SA) is constructed. Secondly, a trilinear interpolation model and a stochastic gradient algorithm are employed to learn the map by iterative updates, which reduces computational complexity and the use of memory. Thirdly, the resulting map generates the optimal CA50 reference (CA50*) under knock constraint, and then a statistical feedback controller tracks CA50* by altering SA. Finally, experimental validations carried out on a six-cylinder SI gasoline engine demonstrate that the presented scheme contributes to faster response of transient operation condition and the on-board learning loop compensates the map drift.