Implementing improved model predictive control of selective catalytic reduction

Oxides of nitrogen (NOx), as emitted by lean-burn direct injection internal combustion engines, have been shown to be harmful to human health, which has resulted in a tightening of road vehicle emissions limits for these pollutants. NOx abatement systems, the most popular of which is selective catalytic reduction (SCR), have been developed and proven effective in production, but increasingly stringent legislation mandates further improvement. In particular, SCR suffers from suboptimal NOx conversion at the low catalyst temperatures occurring in slow moving urban traffic, yet it is precisely these environments where pollutants cause maximal harm. In this thesis, the problem of control of after treatment catalysts is considered, in the light of the significant body of recent research into real-time constrained optimal control, otherwise known as model predictive control (MPC). This paradigm, which has so far seen relatively slow adoption by the automotive industry, promises numerous advantages over classical techniques, including systematic handling of constraints, improved closed loop performance, and ease of tuning and calibration. However, the computational demand of MPC has traditionally been considered insurmountable for real-time application to fast systems. The efficacy of MPC as applied to SCR is examined, focusing particularly on challenging operating conditions, including multivariable control by the addition of a catalyst heater in order to cope with urban traffic conditions. It is demonstrated that SCR can exploit the many benefits offered by MPC. Furthermore, this thesis demonstrates that solving the associated optimal control problem is possible in real-time on a low power automotive grade embedded hardware platform, thereby indicating its feasibility for production SCR control and providing a pathway to wider adoption by industry