MAP Learning and Disturbance Observation based Engine Torque Control for Dynamometer Test Bench

Abstract For the engine-dynamometer test bench, the engine torque output (Te) is controlled by the test bench controller through the manipulation of the engine acceleration pedal position (up). Due to the nonlinearities and uncertainties between up and Te, the time-consuming control parameters scheduling is usually necessary using conventional controllers. In this paper, a composite disturbance observation based torque controller is proposed. The nonlinearity, from up to the torque demand in the engine control unit, is compensated by an adaptive feed forward controller, based on the inverse of a self-learning MAP by stochastic gradient decent. All other nonlinearities and uncertainties are lumped as total disturbance. By estimating the total disturbance using the extended state observer in real-time, the plant is enforced as a first-order system to be easily controlled by a simple proportional controller. The proposed controller is validated in a high-fidelity GT-SUITE simulation model. Results show that average absolute torque tracking error is 1.62N·m over the suburban part in European Transient Cycle without the need of control parameters scheduling.