A Self-Aware Robotic Machining Architecture Based on Physics-Informed Neural Networks

Abstract

Industrial robots have been widely used in various industrial scenarios due to their flexibility and expandability. However, robots are susceptible to instability at high speeds and load conditions due to their low stiffness and dynamic characteristics variation with posture. Enhancing the processing stability of industrial robots through data-driven modeling and physical modeling approaches suffers from different drawbacks, such as solution complexity and weak interpretability. With the emergence of physics-informed neural networks (PINNs), new methodologies can be developed to enhance the self-aware processing of robots. In this paper, typical industry application scenarios and dynamics of robotics as well as PINNs are introduced and analyzed, and a framework and method based on PINNs are proposed to enhance the self-aware operation of industrial robots. This framework and methodology contribute to the researcher's efforts to apply PINNs more intensively to robot operation in the future to improve the stability and intelligence of robot operation.