Neural-Network Feedforward Compensation for Precision Contouring Control of Multi-Axis Motion Systems

Abstract

This paper proposes a neural-network compensation (NNC) strategy for precision contouring motion control of multi-axis motion systems. Firstly, some typical contouring tasks are carried out on a biaxial linear-motor-driven motion system, and the true value of contouring error is obtained by numerical calculation method as the training data of an artificial gated recurrent unit (GRU) neural network. Essentially, the GRU network can be viewed as a data based black-box error model which can capture the characteristics of contouring motion rather accurately. Then, the well trained GRU network can predict contouring error of some tasks those have not been conducted during the training session. Finally, the predicted contouring error is compensated into the reference contour as a kind of feedforward control to improve contouring performance. Comparison between the predicted contouring error and the actual one proves the effectiveness of the proposed GRU neural network. Comparative experiments between NNC and iterative learning control (ILC) validate the excellent nature of the proposed NNC scheme. Actually, NNC is easy to implement and can achieve excellent contouring motion performance as ILC, significantly without need of motion repetition and iteration.