Related Experiment Video
Updated: Jun 29, 2026
Identification of Disease-related Spatial Covariance Patterns using Neuroimaging Data
Published on: June 26, 2013
Interpretable bearing fault diagnosis based on wavelet scattering network, PCA dimensionality reduction and PLUKAN
Jun-Jie Xu1, Mu-Quan Lin2, Han-Cheng Hsiang3
1The Higher Educational Key Laboratory for Flexible Manufacturing Equipment Integration of Fujian Province, Xiamen Institute of Technology, Xiamen, 361021, China.
This study introduces an interpretable bearing fault diagnosis framework using Wavelet Scattering Network (WSN), Principal Component Analysis (PCA), and Piecewise Linear Unit-based Kolmogorov-Arnold Network (PLUKAN). The WSN-PCA-PLUKAN model achieves high accuracy, efficiency, and interpretability for predictive maintenance.
Area of Science:
- Engineering
- Machine Learning
- Signal Processing
Background:
- Conventional deep learning models for bearing fault diagnosis suffer from poor interpretability, weak generalization in small-sample scenarios, and high deployment costs.
- Existing methods often act as
- black boxes
- hindering trust and adoption in industrial settings.
Purpose of the Study:
- To develop an interpretable fault diagnosis framework for bearings that addresses the limitations of conventional deep learning models.
- To enhance diagnostic accuracy, computational efficiency, and model interpretability for industrial predictive maintenance.
Main Methods:
- A three-stage collaborative framework integrating Wavelet Scattering Network (WSN) for robust feature extraction, Principal Component Analysis (PCA) for dimensionality compression, and Piecewise Linear Unit-based Kolmogorov-Arnold Network (PLUKAN) for interpretable classification.
- WSN extracts multi-scale time-frequency scattering features with translation invariance and noise robustness.
- PLUKAN provides structured decision interpretation, overcoming the
- black-box
- nature of traditional models.
Main Results:
- Achieved over 99% average accuracy with an 80% training ratio on multiple datasets.
- Demonstrated strong resilience under extreme few-shot conditions, maintaining competitive accuracy (87.7%-100%) with only 10% of samples for training.
- PLUKAN reduced training and inference times by approximately 50% compared to traditional KAN, meeting industrial edge device requirements.
- Established clear correspondence between feature variations and fault mechanisms through interpretable WSN features and PLUKAN's structure.
Conclusions:
- The proposed WSN-PCA-PLUKAN framework offers a novel, engineering-feasible solution for bearing fault diagnosis.
- The framework successfully integrates diagnostic accuracy, computational efficiency, and interpretability, crucial for industrial predictive maintenance.
- This approach provides a transparent and reliable alternative to conventional deep learning models in critical machinery monitoring.