Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Global-focal Adaptation with Information Separation for Noise-robust Transfer Fault Diagnosis

Junyu Ren, Wensheng Gan, Guangyu Zhang, Wei Zhong, Philip S. Yu

TL;DR

This work tackles the practical problem of fault diagnosis under simultaneous noise and domain shift by introducing ISGFAN, a two-pronged framework that (1) separates fault-relevant information from noise and domain-private cues via an information separation architecture with an improved orthogonality constraint, and (2) performs global-focal domain adaptation to align both marginal and conditional distributions. The global component ensures overall distribution alignment, while the focal component uses class-specific subdomain adaptation with pseudo-labeling and a subdomain attention mechanism to mitigate hard transfer subdomains in unsupervised settings. Empirical results on CWRU, KAIST, and PU datasets demonstrate that ISGFAN achieves state-of-the-art transfer performance under various noise conditions, outperforming strong baselines and maintaining stability across tasks. The work provides a practical, noise-robust approach to cross-domain fault diagnosis with potential impact on industrial reliability by enabling more reliable condition monitoring under noisy, real-world operation. Limitations include reliance on pseudo-label quality for the focal adaptation and the need for careful tuning of the subdomain attention mechanism, suggesting directions for future work in training efficiency and robustness.

Abstract

Existing transfer fault diagnosis methods typically assume either clean data or sufficient domain similarity, which limits their effectiveness in industrial environments where severe noise interference and domain shifts coexist. To address this challenge, we propose an information separation global-focal adversarial network (ISGFAN), a robust framework for cross-domain fault diagnosis under noise conditions. ISGFAN is built on an information separation architecture that integrates adversarial learning with an improved orthogonal loss to decouple domain-invariant fault representation, thereby isolating noise interference and domain-specific characteristics. To further strengthen transfer robustness, ISGFAN employs a global-focal domain-adversarial scheme that constrains both the conditional and marginal distributions of the model. Specifically, the focal domain-adversarial component mitigates category-specific transfer obstacles caused by noise in unsupervised scenarios, while the global domain classifier ensures alignment of the overall distribution. Experiments conducted on three public benchmark datasets demonstrate that the proposed method outperforms other prominent existing approaches, confirming the superiority of the ISGFAN framework. Data and code are available at https://github.com/JYREN-Source/ISGFAN

Global-focal Adaptation with Information Separation for Noise-robust Transfer Fault Diagnosis

TL;DR

This work tackles the practical problem of fault diagnosis under simultaneous noise and domain shift by introducing ISGFAN, a two-pronged framework that (1) separates fault-relevant information from noise and domain-private cues via an information separation architecture with an improved orthogonality constraint, and (2) performs global-focal domain adaptation to align both marginal and conditional distributions. The global component ensures overall distribution alignment, while the focal component uses class-specific subdomain adaptation with pseudo-labeling and a subdomain attention mechanism to mitigate hard transfer subdomains in unsupervised settings. Empirical results on CWRU, KAIST, and PU datasets demonstrate that ISGFAN achieves state-of-the-art transfer performance under various noise conditions, outperforming strong baselines and maintaining stability across tasks. The work provides a practical, noise-robust approach to cross-domain fault diagnosis with potential impact on industrial reliability by enabling more reliable condition monitoring under noisy, real-world operation. Limitations include reliance on pseudo-label quality for the focal adaptation and the need for careful tuning of the subdomain attention mechanism, suggesting directions for future work in training efficiency and robustness.

Abstract

Existing transfer fault diagnosis methods typically assume either clean data or sufficient domain similarity, which limits their effectiveness in industrial environments where severe noise interference and domain shifts coexist. To address this challenge, we propose an information separation global-focal adversarial network (ISGFAN), a robust framework for cross-domain fault diagnosis under noise conditions. ISGFAN is built on an information separation architecture that integrates adversarial learning with an improved orthogonal loss to decouple domain-invariant fault representation, thereby isolating noise interference and domain-specific characteristics. To further strengthen transfer robustness, ISGFAN employs a global-focal domain-adversarial scheme that constrains both the conditional and marginal distributions of the model. Specifically, the focal domain-adversarial component mitigates category-specific transfer obstacles caused by noise in unsupervised scenarios, while the global domain classifier ensures alignment of the overall distribution. Experiments conducted on three public benchmark datasets demonstrate that the proposed method outperforms other prominent existing approaches, confirming the superiority of the ISGFAN framework. Data and code are available at https://github.com/JYREN-Source/ISGFAN
Paper Structure (28 sections, 26 equations, 16 figures, 12 tables, 2 algorithms)

This paper contains 28 sections, 26 equations, 16 figures, 12 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Noise-robust Fault Diagnosis
  4. Transfer Fault Diagnosis
  5. Preliminaries
  6. Problem Setting
  7. Adversarial Domain Adaptation
  8. Proposed Method
  9. Information Separation Framework
  10. Global-Focal Domain Adaptation
  11. Concrete Structure of Model Components
  12. Model Training and Dynamic Loss Weighting
  13. Experiment and Results Analysis
  14. Experimental Setup
  15. Model Training Configuration
  16. ...and 13 more sections

Figures (16)

  • Figure 1: Transfer processes across various situations.
  • Figure 2: The framework of DANN.
  • Figure 3: Conceptual illustration of the proposed method.
  • Figure 4: The framework of information separation global-focal adversarial network.
  • Figure 5: Concrete structure of model components: (a) Fault-relevant feature extraction block; (b) Fault-relevant feature extractor; (c) Fault-irrelevant feature extractor; (d) Fault-irrelevant feature extraction block; (e) Generic structure of classifiers and discriminators; (f) Multi-layer perceptron; (g) Subdomain classifiers; (h) Decoder.
  • ...and 11 more figures