Multivariable Combustion Control for Engines Operating Near High-Variability Limits

Abstract

This thesis contributes strategies for the closed-loop learning and control of stochastic combustion behavior in a compression-ignition internal combustion engine operating near high-variability limits. Engines of the future must work safely and efficiently in diverse ambient conditions with a wide range of sustainable non-petroleum-based fuels that may possess varying or uncertain properties. Because traditional feedforward control calibrations are only tuned to a finite range of operating conditions, feedback control schemes supported by in-cylinder pressure sensing are needed. The feedback-based strategies in this thesis solve the unique challenges of controlling combustion under extremely adverse conditions induced by low pressure-temperature inlet conditions or fuels with low cetane number (CN) that have poor ignition behavior.

Two strategies are presented for tracking combustion phasing to a setpoint on a cycle-to-cycle basis by coordinating fuel start of injection (SOI) with power supplied to an electrically heated in-cylinder ignition assist (IA) device. The controllers achieve rapid tracking of combustion phasing step commands, supply IA power in accordance with actuator range and rate constraints, and reduce combustion variability in the feedback loop using Kalman filtering of feedback measurements. The cycle-to-cycle setpoint tracking control loop is then augmented with an online risk-aware combustion learning algorithm. In response to a fuel change disturbance, the combined learning and control scheme tactically coordinates SOI and IA to learn new edge limits, maximize mean-value efficiency, and mitigate undesirable high in-cylinder pressure rise rates to within a tolerable probabilistic threshold. The learning and control schemes are validated in simulation using nonlinear statistical virtual engines that replicate both transient and steady-state stochastic combustion behavior as a function of SOI, IA power, and either thermodynamic state or fuel CN. These virtual engines were tuned with data from engine experiments conducted at low pressure-temperature inlet and low cetane fueling conditions that induced highly variable combustion behavior. Ultimately, this thesis advances closed-loop combustion control by demonstrating how to leverage novel ignition assisting actuators, manage high cycle-to-cycle variability, and navigate unexpected shifts in the optimal operating region and probabilistic limits of combustion.