Stochastic Analysis and Control of EGR-Diluted Combustion in Spark Ignition Engines at Nominal and Misfire-Limited Conditions

Abstract

Worldwide regulations on greenhouse gas emissions demand a reduction in fuel consumption from the transportation sector. This reduction requires incremental improvements in engine and powertrain efficiency. Feedback combustion control under diluted conditions with exhaust gas recirculation (EGR) has the potential to improve the overall efficiency of spark-ignition engines by optimizing combustion efficiency, reducing heat transfer losses, and reducing pumping losses at medium loads. This control problem requires the coordinated action of the EGR valve and the spark advance. However, cycle-to-cycle variability in the combustion process limits the closed-loop system performance. Moreover, the input-to-output coupling between the actuators and measured combustion features need to be addressed in the control design to avoid undesired combustion events such as knock, partially burned cycles, and misfires. Therefore, the combustion control problem at high EGR-diluted conditions is a constrained multivariable stochastic control problem. This dissertation focuses on the control of the spark advance and the EGR valve in order to maximize the EGR benefits while maintaining stable combustion during steady state and load transients.

For a fixed engine speed/load condition, a two-input two-output discrete-time dynamic system was derived from system identification in order to use model-based control techniques. In particular, a linear quadratic Gaussian (LQG) controller was designed and experimentally tested for controlling spark and EGR valve. Such a controller was able to achieve an optimal combustion shape that maximizes EGR benefits and proved to be superior compared to traditional proportional-integral (PI) control strategies. An analytic solution for the amount of variability that the LQG controller contributes during closed-loop operation was derived, which can be used to modify the combustion targets to avoid misfire events. Given that sporadic misfires can occur when the control targets high levels of EGR, a stochastic controller based on the likelihood ratio test has been proposed to adjust the likelihood of misfires.

When the engine speed is fixed and the load demand is controlled by the driver, the feedback combustion controller needs to react to such disturbance and maintain an optimal phasing. A physics-based model derived from manifold filling dynamics was coupled with a simple combustion model to formulate a three-input two-output dynamic system that considers not only the impact of the EGR valve and spark advance on the combustion, but also considers throttle tip-in and tip-out commands. The retuned LQG controller was experimentally tested and successfully maintained optimal phasing and maximized EGR levels during tip-in commands. However, during throttle tip-outs the system transitions through conditions where misfires occur. An explicit reference governor was designed to slow down the tip-out commands in order to avoid fast transitions that drive the system over the misfire limit. Given the inability to model misfires accurately, the reference governor was enhanced with model-free learning which enabled it to avoid misfires over time. Experimental results showed that successful misfire avoidance can be achieved in exchange for a slower tip-out response. It is suggested that such combustion control strategies can be paired with modern mild or full hybrid powertrain architectures to fully utilize the advantages of combustion control at high dilution levels.