一种不平衡数据样本下的旋转机械故障诊断方法
A time-frequency consistency self-supervised learning method based on pseudo-twin structures was used to construct a fault diagnosis model for rotating machinery. This method solved the problem of fault diagnosis under unbalanced data, achieved high-precision fault detection and classification, and improved the model's generalization ability by extracting general features from normal data.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HANGZHOU INTERNATIONAL INNOVATION INSTITUTE OF BEIHANG UNIVERSITY
- Filing Date
- 2026-03-25
- Publication Date
- 2026-07-17
AI Technical Summary
Existing methods for diagnosing rotating machinery faults struggle to effectively utilize large amounts of normal data under unbalanced data samples, lack a unified model framework, resulting in poor generalization ability for rare faults, and existing graph comparison learning methods rely on manual trial and error and expensive domain knowledge.
A time-frequency consistency self-supervised learning method with pseudo-twin structure is adopted. By constructing a time-frequency graph joint feature extraction model through a graph comparison learning network with nested pseudo-twin structure, a self-supervised pre-training method is performed using normal samples. Combined with the topological relationship of time-frequency signals, high-precision fault detection and classification are achieved.
Achieving high-precision fault detection and classification of rotating machinery with few samples avoids manual trial and error and the need for expensive domain knowledge. It makes full use of normal data to extract general features and improves the model's generalization ability.
Smart Images
Figure CN122413005A_ABST