Manipulator joint fault localization for intelligent flexible manufacturing based on reinforcement learning and robot dynamics

Abstract

This article proposes a new method to address the challenge of remote monitoring in intelligent flexible manufacturing systems. Specifically, we propose a multi working condition fault localization algorithm for robotic arms, which eliminates the need for additional sensors and is based on the classic sliding window algorithm. We use reinforcement learning technology to learn detection parameter debugging experience under different working conditions, and combine the dynamics of the robot to achieve fault detection and fault source localization in a flexible environment. Through the robot's own programmable logic controller system, the remote monitoring system can sense the operating status of each link. To evaluate the effectiveness of our proposed method, we conducted experimental equipment simulations and real-world industrial operations. The results show that under multiple operating conditions, the accuracy of fault detection reaches 86%, and the accuracy of localization reaches 81.35%. The deviation of results under different robot operating conditions is significantly lower than other algorithms. This study explores the potential and implementation approaches of reinforcement learning in intelligent manufacturing systems, with a particular focus on applications in flexible scenarios. These findings reveal the prospects of reinforcement learning technology in improving the sustainable operation of intelligent manufacturing systems.