Toward Real-Time Automotive Model Predictive Control: A Perspective from a Diesel Air Path Control Development

To meet increasingly stringent fuel economy and emissions regulations, automotive powertrain architectures are becoming increasingly sophisticated, e.g., through downsizing and electrification, with increasing numbers of actuators and interacting dynamics. Simultaneously, the corresponding control systems are becoming more complex and the degree of difficulty involved in developing them using traditional methods is increasing. Model Predictive Control (MPC) has long been identified as a promising technique to control complex power-train dynamics under performance, reliability, and regulatory constraints. However, despite an abundance of research in applying MPC to powertrain control, the computational burden of these MPC strategies is still too great for widespread production adoption of MPC on low cost embedded control units (ECU). This paper summarizes our path toward a real-time-on-ECU model predictive controller in the context of a Diesel air path control development and comments on model accuracy, optimal control problem formulation, solver accuracy, performance, robustness, and practicality.