Bearing fault detection by using graph autoencoder and ensemble learning.

Area of Science:

• Mechanical Engineering
• Data Science
• Artificial Intelligence

Background:

• Bearing fault diagnosis is vital for equipment reliability and maintenance cost reduction.
• Existing methods struggle to differentiate normal and faulty machine conditions, causing unreliable detection.
• Advanced diagnostic methodologies are needed to overcome current limitations.

Purpose of the Study:

• To propose a novel bearing fault detection approach.
• To enhance the accuracy and stability of fault detection in machinery.
• To address the limitations of existing signal discrimination techniques.

Main Methods:

• A novel stochasticity-based compositional method to convert Euclidean data into graph format.
• Application of graph neural networks (GNNs) for processing graph-structured data.
• Integration of feature fusion and a new ensemble learning strategy for outlier detection.

Main Results:

• Successfully developed a robust method for bearing fault detection.
• Demonstrated improved accuracy in discriminating between normal and faulty machine states.
• The proposed ensemble learning strategy enhances outlier detection capabilities.

Conclusions:

• The novel approach significantly advances bearing fault diagnosis accuracy.
• This research provides a pivotal contribution to enhancing diagnostic methodologies for machinery.
• The integration of GNNs and ensemble learning offers a promising direction for future research.