Real-Time Constrained Optimal Control of Diesel Engine Airpath

Abstract

The airpath control for a diesel engine is a multi-variable problem with nonlinear dynamicsand operation constraints. Using an experimentally validated model of a EURO 6 engine, thisthesis proposes three airpath control strategies that tackle the control problem at differentlevels. The proposed control strategies are realised using Model Predictive Control (MPC),a control methodology that offers an optimal control framework to systematically considermulti-variable dynamics and constraints. The first control strategy is a nonlinear MPC thatsolves a tracking problem where the desired set points, including intake manifold pressure andcylinder oxygen concentration, are given. Compared to a production-line controller, the controller achieves 10% and 18% performance improvements in tracking intake manifold pressureand cylinder oxygen concentration, respectively. The second control strategy is a nonlineareconomic MPC (eMPC) that determines optimal set points online to simultaneously reduceengine emissions, fuel economy and improve the tracking of torque. It exploits the highlevel performance objectives to calculate the Exhaust Gas Recirculation (EGR) and VariableNozzle Turbocharger (VNT) valve positions. Compared to a production-line controller, theeMPC results in 1.24%, 4%, 3% improvements in fuel economy, NOx and soot emissions,respectively. Building on the eMPC, a look-ahead eMPC (LAeMPC) is developed to exploita horizon of known future operating points, known as look-ahead information. The LAeMPCdetermines the control that exploits 2 seconds of look-ahead information and shows further5% and 11% improvement of NOx and soot without compromising fuel economy. Finally, thereal-time capabilities of all the control strategies are demonstrated on a System-on-Module(SoM) in a Hardware-in-the-Loop (HIL) environment. The controllers are evaluated for theWorldwide harmonised Light vehicles Test Cycles (WLTC). The NMPC takes up to 2.9 mson a 480 MHz ARM processor. The eMPC and LAeMPC take up to 9.6 and 8.9 ms on a1200 MHz ARM processor, respectively.

Date of Award	May 2022
Original language	English
Awarding Institution	Coventry University
Supervisor	Olivier haas (Supervisor), Arash Moradinegade Dizqah (Supervisor), Martin Herreros (Supervisor) & Joschka Schaub (Supervisor)