Current Only-Based Fault Diagnosis Method for Industrial Robot Control Cables.

Area of Science:

• Industrial Automation and Control Systems
• Machine Learning for Fault Diagnosis
• Electrical Engineering and Signal Processing

Background:

• Deep learning methods are increasingly used for fault diagnosis in factory automation due to automatic feature extraction.
• Novelty detection is crucial for imbalanced datasets and faults difficult to replicate in labs.
• Control cable integrity is vital for reliable industrial robot operation.

Purpose of the Study:

• To propose a novelty detection-based soft fault diagnosis method for control cables.
• To utilize only current signals for fault detection, simplifying data acquisition.
• To enable accurate diagnosis under imbalanced data and environmental disturbances.

Main Methods:

• A novelty detection algorithm using three-phase currents to calculate current sums and ratios.
• An autoencoder architecture for novelty detection and anomaly score calculation.
• A moving average filter and defined threshold for robust soft fault diagnosis.

Main Results:

• The proposed method accurately diagnoses soft faults in control cables across 11 fault scenarios.
• Faults were induced by gradually damaging cable conductors and insulators.
• Effective diagnosis was achieved under various operating conditions and fault severities.

Conclusions:

• The novelty detection approach effectively diagnoses soft faults in control cables using current data.
• The method demonstrates robustness against environmental disturbances and data imbalance.
• This technique offers a reliable solution for maintaining industrial robot operational integrity.