A method and system for fault diagnosis of rolling bearings based on dual-stream gated cross-modal fusion

The problem of extracting fault features of rolling bearings in high-noise environments was solved by using a dual-stream gated cross-modal fusion network (DS-GCMAF), which achieves high robustness and high accuracy in fault diagnosis and is suitable for real-time industrial monitoring.

CN122310218APending Publication Date: 2026-06-30GUANGDONG UNIV OF PETROCHEMICAL TECH
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
GUANGDONG UNIV OF PETROCHEMICAL TECH
Filing Date
2026-04-03
Publication Date
2026-06-30

AI Technical Summary

Technical Problem

Existing technologies struggle to effectively extract rolling bearing fault features in high-noise environments, and the reliability of different modal features in multimodal fusion methods fluctuates greatly, leading to noise dispersion effects and negative migration phenomena, which affect the accuracy of fault severity determination.

Method used

A dual-stream gated cross-modal fusion network (DS-GCMAF) is adopted. Through one-dimensional time domain and two-dimensional time-frequency feature extraction branches, combined with a cross-modal multi-head attention fusion module and an adaptive decision fusion module, the fusion strategy is dynamically adjusted to suppress damaged modalities and utilize more robust features for cross-calibration.

Benefits of technology

It significantly improves the accuracy and robustness of fault diagnosis in high-noise environments, ensuring high-precision diagnosis under extreme noise conditions, and has high computational efficiency, making it suitable for real-time industrial monitoring.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122310218A_ABST
    Figure CN122310218A_ABST
Patent Text Reader

Abstract

This invention discloses a rolling bearing fault diagnosis method based on a dual-stream gated cross-modal fusion network, belonging to the field of industrial fault diagnosis technology. The method first slices and performs short-time Fourier transform on the acquired bearing vibration signal to construct a dual-modal sample library containing one-dimensional time-domain and two-dimensional time-frequency images. Second, a dual-stream gated cross-modal attention fusion network (DS-GCMAF) is constructed, using parallel branches to extract deep features in the time and frequency domains respectively. Subsequently, feature depth interaction is performed through the cross-modal attention fusion module, and a quality-aware adaptive gating mechanism is used to automatically suppress contaminated modes under strong noise, with highly robust features as anchor points for cross-calibration. Finally, the fault type and severity are output through an adaptive decision fusion module. This invention effectively overcomes the limitations of single-modal analysis under complex noise and the negative transfer problem caused by traditional feature splicing, significantly improving the accuracy and robustness of fault diagnosis under extreme operating conditions.
Need to check novelty before this filing date? Find Prior Art