Using a learning controller to achieve accurate linear motor motion control

The development and experimental implementation of a learning controller for achieving accurate (on the order of micrometers) linear motor motion control of a rigid mass is presented. The learning controller employs (1) neural networks with local basis functions to approximate system nonlinearities, and thus compensate for them, and (2) a linear control component consisting of both feedback and feedforward terms. The significant nonlinearities present in our system are established to be friction, cogging, and torque ripple. It is assumed that detailed knowledge of these nonlinearities is not available. Experimental results demonstrate that the proposed learning controller is able to significantly reduce trajectory tracking errors when compared to a standard controller.