Spectrum-aware prototypical contrastive learning for cross-domain fault diagnosis
Authors:
- 1 March 2026
- journal article
- Published by IOP Publishing in Engineering Research Express
- Vol. 8 (5) , 055319
Abstract
Distribution shifts induced by varying operating conditions severely impede the cross-domain generalization of machinery fault diagnosis models. Existing methods often fail to fully exploit frequency-domain features and suffer from complex hyperparameter tuning. To address these limitations, this paper proposes a Spectrum-Aware Prototypical Contrastive Learning (SAPCL) method. SAPCL introduces a dual-stream architecture that adaptively fuses temporal and spectral features, effectively overcoming feature distortion caused by speed or load variations. Furthermore, to clarify ambiguous decision boundaries, we design a momentum-updated prototypical contrastive learning mechanism that enforces intra-class compactness. Crucially, we propose a unified optimization objective controlled by a single hyperparameter, significantly reducing the complexity of hyperparameter tuning. Extensive experiments on CWRU and JNU datasets demonstrate that SAPCL achieves an average accuracy of 99.30%, outperforming state-of-the-art methods, particularly in challenging cross-speed scenarios. Ablation studies confirm that the spectrum-aware fusion contributes an 11.10% performance gain, validating the necessity of frequency-domain information.Keywords
Funding Information
- Anhui Provincial Natural Science Foundation Energy Internet Fund Project (2008085UD14)
This publication has 30 references indexed in Scilit:
- A Dual-Stream Gated Attention Network for Fault Diagnosis of Drill Bit Using time and Frequency Domain-based Vibration FeaturesJournal of Vibration Engineering & Technologies, 2025
- A systematic review on advancement and challenges in multi-fault diagnosis of rotating machinesEngineering Applications of Artificial Intelligence, 2025
- Cross-sensor contrastive learning-based pre-training for machinery fault diagnosis under sample-limited conditionsKnowledge-Based Systems, 2025
- Time- and frequency-domain fusion for source-free adaptation fault diagnosisInformation Fusion, 2025
- Multi-source domain adaptation using diffusion denoising for bearing fault diagnosis under variable working conditionsKnowledge-Based Systems, 2024
- An auto-regulated universal domain adaptation network for uncertain diagnostic scenarios of rotating machineryExpert Systems with Applications, 2024
- Time-frequency supervised contrastive learning via pseudo-labeling: An unsupervised domain adaptation network for rolling bearing fault diagnosis under time-varying speedsAdvanced Engineering Informatics, 2023
- Domain adaptation network base on contrastive learning for bearings fault diagnosis under variable working conditionsExpert Systems with Applications, 2022
- Deep learning algorithms for rotating machinery intelligent diagnosis: An open source benchmark studyISA Transactions, 2020
- Deep CORAL: Correlation Alignment for Deep Domain Adaptation2016