Neural network-based iterative learning control of a piezo-driven nanopositioning stage

Abstract

The piezo-driven nanopositioning stage (PNS) is a key device to provide fast and precise motions for applications such as micromanipulation, microfabrication, and microscopy scanning. However, inherent nonlinearities associated with system perturbations bring difficulties to controller design. Regarding repetitive tasks for a PNS, existing control schemes are mainly dedicated to model inversion-based iterative learning control (ILC), which relies heavily on model accuracy. In this paper, a novel online identification and control scheme named neural network-based ILC (NN-ILC) is proposed for repetitive tracking of the PNS. The ILC scheme reduces repetitive errors due to the linear dynamics and invariable disturbance during each iteration. Neural networks are integrated into the ILC scheme to minimize the residual non-repetitive errors resulting from unknown nonlinear dynamics and model perturbations. Convergence results in both the time and iteration domains are demonstrated according to the Lyapunov stability theory. Comprehensive experiments of sinusoidal and triangular tracking references with different frequencies (5∼20 Hz) and different peak-to-peak amplitudes (5∼20μm) are conducted on a real-time control testbed. Results show that the root mean square error of the proposed NN-ILC for 20 μm tracking cases is improved by up to 37% from feedback proportional-derivative (PD) control with neural networks and by up to 20% from feedforward PD-type ILC.