Cycle Adaptive Feedforward Approach Controllers for an Electromagnetic Valve Actuator

An electromagnetic valvetrain for an internal combustion engine can improve the engine thermal efficiency but requires soft landing control to avoid excessive wear and acoustic noise. To simplify the soft landing problem, the valve control algorithm is partitioned into approach and landing controllers. The landing control is responsible for the last part of the 8 mm valve travel while the approach control is responsible for the rest. This paper focuses on the approach control. The goal of the approach control is to achieve an end state that sets constant initial conditions for the landing control. In addition to an identified system model, information based on previous valve events is utilized to take advantage of the repetitive nature of the valve opening/closing. Nonlinear iterative learning, terminal iterative learning, and Nelder Mead direct search algorithms are three cyclic adaptive feedforward approach controllers that are tested in simulation for automotive electromagnetic valves. These results are compared but the emphasis of this paper is on the Nelder Mead approach which works well both in simulation and experiment for disturbances that are slow compared to the valve travel time.