Fault Diagnosis Method for Human Coexistence Robots Based on Convolutional Neural Networks Using Time-Series Data...

Area of Science:

• Robotics
• Artificial Intelligence
• Machine Learning

Background:

• Industrial environments increasingly utilize human coexistence robots (HCRs), necessitating robust safety measures.
• Proactive fault diagnosis is critical for preventing safety incidents in HCRs.
• Limited post-failure data poses a challenge for training diagnostic models.

Purpose of the Study:

• To develop and evaluate novel fault diagnosis methods for HCR driving modules.
• To address the challenge of limited fault data using generative models.
• To identify optimal image encoding and convolutional neural network (CNN) architectures for time-series vibration data.

Main Methods:

• Collected time-series vibration data from HCR driving module durability tests.
• Employed Wasserstein generative adversarial networks with gradient penalty (WGAN-GP) to augment limited fault data.
• Converted time-series data into images using recurrence plot, Gramian angular field, Markov transition field, spectrogram, and scalogram.
• Applied and compared four CNN models (VGGNet, GoogleNet, ResNet, DenseNet) for fault classification.

Main Results:

• WGAN-GP significantly improved diagnostic accuracy by generating synthetic fault data.
• Spectrogram image encoding and the DenseNet CNN model demonstrated superior performance.
• The proposed methods achieved high accuracy in diagnosing faults in HCR driving modules.

Conclusions:

• The integration of WGAN-GP for data augmentation and spectrograms with DenseNet offers a powerful approach for robot fault diagnosis.
• This methodology enhances the safety and reliability of industrial robots, particularly HCRs.
• The findings provide a framework for selecting effective image encoding and CNN models for time-series fault analysis.