Design of an Active Fault-Tolerant Control Strategy for a Robotic Manipulator Using Physics-Informed Neural Networks

Abstract

Robotic arms equipped with sensors and actuators are utilized in various industrial applications. Since sensors have a substantial impact on how reliably robotic manipulators can operate, related faults have the potential to impact system performance and stability. This study proposes an active fault-tolerant control system (AFTCS) for a reduced-order three-degree-of-freedom PUMA560 robotic manipulator using physics-informed neural networks (PINNs). By reconfiguring joint states with the PINNs functioning as an estimator, the suggested approach enables uninterrupted operation when any sensor fault occurs. The method is implemented and validated for robotic arm feedback joint angle sensors using MATLAB/Simulink, combining physics-informed learning with fault-tolerant control (FTC) to make systems more resilient. MATLAB/Simulink is used for the implementation and validation of joint angle sensors. The AFTCS maintains system stability and tracking even in the presence of a sensor fault. The findings from the simulation show that the technique stabilizes the system, keeps monitoring performance, and deals with sensor failures effectively.