Multi-Representation Domain Adaptation Network with Duplex Adversarial Learning for Hot-Rolling Mill Fault Diagnosis
Rongrong Peng1,2, Xingzhong Zhang2, Peiming Shi3
1Nonlinear Dynamics and Application Research Center, Nanchang Institute of Science and Technology, Nanchang 330108, China.
Entropy (Basel, Switzerland)
|January 21, 2023
Summary
This study introduces a novel multi-representation domain adaptation network for hot rolling mill fault diagnosis. The method enhances diagnostic accuracy under variable conditions by using duplex adversarial learning for robust feature transfer.
Area of Science:
- Industrial Engineering
- Machine Learning
- Mechanical Engineering
Background:
- Steel rolling manufacturing involves complex, variable conditions challenging equipment fault diagnosis.
- Existing transfer learning methods for fault diagnosis have limitations in handling diverse operating conditions due to single-representation alignment.
- This can lead to incomplete knowledge transfer and imprecise fault classification boundaries.
Purpose of the Study:
- To propose a novel multi-representation domain adaptation network with duplex adversarial learning for enhanced hot rolling mill fault diagnosis.
- To improve the accuracy and robustness of fault diagnosis under variable working conditions.
- To address the limitations of single-representation transfer learning methods.
Main Methods:
- Designed a multi-representation network to extract equipment status information from multiple perspectives.
- Employed domain adversarial strategy for distribution alignment across source and target domains, learning domain-invariant features.
- Utilized maximum classifier discrepancy adversarial algorithm to generate target features close to source support, forming robust decision boundaries.
Main Results:
- The proposed method achieved high average diagnostic rates of 99.15% for rolling mill gearboxes and 99.40% for bearings.
- Demonstrated significant improvement over the most competitive existing methods, with increases of 2.19% and 1.93% respectively.
- Effectively realized fault diagnosis of rolling mill equipment under variable working conditions.
Conclusions:
- The multi-representation domain adaptation network with duplex adversarial learning is effective for hot rolling mill fault diagnosis.
- The approach successfully overcomes challenges posed by variable working conditions.
- This method offers a robust solution for improving the reliability and accuracy of industrial equipment monitoring.
