A simulation data-driven semi-supervised framework based on MK-KNN graph and ESSGAT for bearing fault diagnosis
Yuyan Li1, Tiantian Wang2, Jingsong Xie1
1College of Traffic and Transportation Engineering, Central South University, Changsha 410075, China.
ISA Transactions
|October 4, 2024
Summary
This study introduces a semi-supervised framework for intelligent fault diagnosis using unlabeled data. The novel approach enhances diagnostic accuracy by effectively utilizing simulation and real-world data, overcoming labeling challenges.
Area of Science:
- Mechanical Engineering
- Artificial Intelligence
- Data Science
Background:
- Supervised intelligent fault diagnosis requires extensive labeled data, which is costly and time-consuming to acquire.
- The challenge of fault diagnosis using unlabeled data necessitates advanced methodologies.
Purpose of the Study:
- To propose a novel simulation data-driven semi-supervised framework for intelligent fault diagnosis.
- To address the limitations of traditional supervised methods by leveraging unlabeled data.
Main Methods:
- A multi-kernel K-nearest neighbor (MK-KNN) approach was developed to establish neighborhood relationships between simulation and real data, enhancing graph robustness.
- An edge self-supervised graph attention network (ESSGAT) was designed to predict edge presence and emphasize critical neighboring nodes within the MK-KNN graph.
Main Results:
- The proposed MK-KNN and ESSGAT framework demonstrated superior diagnostic performance on bearing and high-speed train axle box bearing datasets.
- Comparative analysis showed improved results over existing state-of-the-art graph construction methods and graph convolutional networks.
Conclusions:
- The developed semi-supervised framework effectively addresses the challenge of fault diagnosis with limited labeled data.
- This approach offers a robust and accurate solution for intelligent fault diagnosis in mechanical systems.
