Leveraging Model Predictive Control As A Calibration Method To Develop Implementable Vehicle Dynamics Controls

As on-board computing power in automotive ECUs grows, the possibility of running an online Model Predictive Control (MPC) algorithm is becoming a reality. However, there still exists a gap between developmental hardware used for MPC development and production-grade hardware. Nevertheless, engineers can use MPC to investigate the most optimal way of obtaining the desired control output. This paper uses the front/rear wheel torque allocation problem in an electric all-wheel-drive (eAWD) vehicle as a case study to investigate how MPC can be leveraged as a calibration method to develop an implementable torque split control system with a good initial calibration. The eAWD powertrain architecture is modeled and integrated with a high fidelity CarSim vehicle dynamics model. An idealized MPC controller is developed and used in several use cases with the CarSim vehicle in the loop to collect data on how the MPC controller obtains the optimal wheel torque distribution. The collected data is then used to develop an implementable control system which is easily calibrateable. The performance of the implementable control system is then compared to the idealized MPC controller in the simulation environment and it is shown that the implementable controller obtains similar performance to the idealized MPC controller.