Physics-Guided Neural Networks for Feedforward Control: An Orthogonal Projection-Based Approach

TL;DR

This paper introduces a physics-guided neural network approach for feedforward control that effectively handles unknown nonlinear dynamics by orthogonal projection, improving performance and generalization.

Contribution

It proposes a novel orthogonal projection-based neural network framework that enhances model identifiability and control performance for systems with unknown nonlinear dynamics.

Findings

Improved control performance on nonlinear friction system

Enhanced model identifiability and generalization

Validated effectiveness through experimental results

Abstract

Unknown nonlinear dynamics can limit the performance of model-based feedforward control. The aim of this paper is to develop a feedforward control framework for systems with unknown, typically nonlinear, dynamics. To address the unknown dynamics, a physics-based feedforward model is complemented by a neural network. The neural network output in the subspace of the model is penalized through orthogonal projection. This results in uniquely identifiable model coefficients, enabling both increased performance and good generalization. The feedforward control framework is validated on a representative system with performance limiting nonlinear friction characteristics.