Sensor and Actuator Fault Diagnosis for Robot Joint Based on Deep CNN

Area of Science:

• Robotics
• Artificial Intelligence
• Machine Learning

Background:

• Robot joint faults, including sensor and actuator issues like gain error, offset error, and malfunction, pose significant challenges to operational reliability.
• Accurate and efficient fault diagnosis is crucial for maintaining robotic system performance and safety.

Purpose of the Study:

• To propose and validate a data-driven fault diagnosis method for robot joints utilizing a deep convolutional neural network (DCNN).
• To enhance fault recognition accuracy and minimize model training time compared to existing methods.

Main Methods:

• A deep convolutional neural network (DCNN) was developed to process fused sensor and actuator data, enabling a unified fault description.
• The DCNN was trained to extract characteristic features from merged data for differentiating various fault types.
• The proposed DCNN model's effectiveness was benchmarked against Support Vector Machine (SVM), Artificial Neural Network (ANN), LeNet-5 Convolutional Neural Network (CNN), and Long-Term Memory Network (LTMN).

Main Results:

• The DCNN-based fault diagnosis method demonstrated superior fault recognition accuracy.
• The DCNN model required significantly less training time compared to the other investigated methods.
• The DCNN effectively learned discriminative features from fused sensor and actuator data for robust fault identification.

Conclusions:

• The proposed DCNN method offers a highly accurate and efficient solution for diagnosing sensor and actuator faults in robot joints.
• The DCNN's ability to learn from fused data and its reduced training time make it a promising approach for real-world robotic applications.