Neural-network-based robust fault diagnosis in robotic systems

Area of Science:

• Robotics
• Control Systems
• Artificial Intelligence

Background:

• Fault diagnosis is critical for modern robotic systems.
• Model-based analytical redundancy is a common approach for robotic manipulator fault diagnosis.
• Modeling uncertainties pose a significant challenge to the performance of fault diagnosis schemes.

Purpose of the Study:

• To investigate fault diagnosis in rigid-link robotic manipulators considering modeling uncertainties.
• To develop a robust fault diagnosis scheme using a learning architecture.
• To rigorously establish the robustness and stability of the proposed fault diagnosis scheme.

Main Methods:

• Utilizing a learning architecture with sigmoidal neural networks.
• Monitoring the robotic system for off-nominal behavior indicative of faults.
• Rigorously establishing robustness and stability properties of the fault diagnosis scheme.

Main Results:

• The proposed neural-network-based scheme effectively detects and accommodates faults.
• The fault diagnosis scheme demonstrates robustness against modeling uncertainties.
• Simulation examples confirm the scheme's ability to handle faults in a two-link robotic manipulator.

Conclusions:

• A robust fault diagnosis scheme using neural networks is effective for robotic manipulators.
• The developed method addresses the challenge of modeling uncertainties in fault diagnosis.
• The approach enhances the reliability and operational safety of robotic systems.