Neural network based time-optimal control of a magnetically levitated precision positioning system

This paper describes an application of artificial neural networks to the problem of time-optimal control of a magnetically levitated platen. The system of interest is a candidate technology for advanced photolithography machines used in the manufacturing of integrated circuits. The nonlinearities associated with magnetic levitation actuators preclude the direct application of classical timeoptimal control methodologies for determining optimal restto-rest maneuver strategies. Instead, a computer simulation of the platen system is manipulated to provide a training set for an artificial neural network. The trained network provides optimal switching times for conducting one dimensional rest-to-rest maneuvers of the platen that incorporate the full nonlinear effects of the magnetic levitation actuators. Sample problems illustrate the effectiveness of the neural network based control as compared to traditional proportional-derivative control.