Intelligent fault identification for industrial automation system via multi-scale convolutional generative...

Area of Science:

• Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Rolling bearings are critical components in industrial automation systems.
• Intelligent fault identification is essential for ensuring system stability.
• A key challenge is the need for extensive labeled data for model training.

Purpose of the Study:

• To address the data scarcity issue in bearing fault identification.
• To develop a semi-supervised model leveraging both labeled and unlabeled data.
• To improve the accuracy and efficiency of intelligent bearing fault detection.

Main Methods:

• Proposed a semi-supervised multi-scale convolutional generative adversarial network (GAN).
• Utilized a one-dimensional multi-scale convolutional neural network as the discriminator.
• Employed a multi-scale deconvolutional neural network as the generator.
• Trained the model using an adversarial process with partially labeled and sufficient unlabeled samples.

Main Results:

• The semi-supervised GAN achieved high classification accuracy on three datasets (100%, 99.28%, 96.58%).
• Demonstrated effective bearing fault detection with limited labeled samples.
• Validated the model's satisfactory performance in data-scarce scenarios.

Conclusions:

• The proposed semi-supervised GAN is a viable solution for intelligent bearing fault identification.
• Leveraging unlabeled data significantly enhances model performance with limited labeled samples.
• The method offers a robust approach for maintaining industrial automation system reliability.