Supervised Machine-Learning Methodology for Industrial Robot Positional Health Using Artificial Neural Networks,...

Area of Science:

• Robotics and Industrial Automation
• Machine Learning for Predictive Maintenance

Background:

• Robotic systems are crucial for repetitive industrial tasks requiring high positional accuracy.
• Degradation in robot positional accuracy leads to significant resource loss.
• Current predictive maintenance methods using external sensors are complex for industrial settings.

Purpose of the Study:

• To propose an effective and simpler method for detecting positional degradation in robot joints.
• To analyze actuator current signals for early fault diagnosis.

Main Methods:

• Utilized discrete wavelet transform for signal processing.
• Applied nonlinear indices and principal component analysis for feature extraction.
• Employed artificial neural networks for classification of positional deviations.

Main Results:

• The proposed methodology accurately detects robot positional degradation using actuator current signals.
• Achieved 100% classification accuracy for robot positional degradation.
• Demonstrated the effectiveness of analyzing current signals for health monitoring.

Conclusions:

• Early detection of positional degradation is feasible by analyzing actuator currents.
• The developed method facilitates timely implementation of prognosis and health management (PHM) strategies.
• This approach helps prevent losses in manufacturing processes by ensuring robot accuracy.