Control- Relevant Neural Networks for Intelligent Motion Feedforward

Abstract

Neural networks have large potential for motion feedforward because of their ability to approximate a wide range of functions. The aim of this paper is to develop a systematic framework for application of neural networks to motion feedforward, that leads to an intelligent motion feedforward approach in the sense that it achieves both flexibility for varying references and high performance. Iterative learning control is used to generate training data, and a control-relevant performance function is introduced. Non-causal feedforward is enabled through two network configurations that enable respectively finite and infinite preview. The approach is experimentally validated on an industrial flatbed printer.