国外文献 > 1.轴承 > 7. 智能方法与计算 > 机器学习
关键词搜索：
| 标 题： |
A Deep Correlation Feature Extraction Network: Intelligent Description of Bearing Fault Knowledge for Zero-Sample Learning
深度相关特征提取网络：用于零样本学习的轴承故障知识智能描述 |
| 作 者： |
Jinbiao Tan
Jiafu Wan
Hu Cai
Xiaowei Chen
Baotong Chen |
| 关键词： |
Deep learning [深度学习]
Zero-sample [ 零样本]
Fault attribute description [ 故障属性描述]
Fault diagnosis [ 故障诊断]
Signal processing knowledge [ 信号处理知识]
Zero-shot learning [ 零样本学习] |
| 出版年： |
2024 |
| 文集名： |
Knowledge Science, Engineering and Management
知识科学、工程与管理 |
| 会议名： |
17th International Conference on Knowledge Science, Engineering and Management (KSEM 2024)
第17届知识科学、工程与管理国际会议（KSEM 2024） |
| 届 次： |
17th |
| 摘 要： |
In zero-shot learning for bearing fault diagnosis, it is crucial to provide an accurate description of fault attributes. However, the existing fault attribute descriptions depend on manual definitions, which is labor-intensive and may not be very accurate. Hence, there is an urgent need to develop an automated system for adaptive attribute description. To tackle this issue, this paper proposes a domain knowledge-driven intelligent description method for bearing fault attributes, that is, a deep correlation feature extraction network for automatic definition of attributes in zero-shot fault diagnosis tasks. Fault attribute descriptions are generated by extracting periodic components of deep features using correlation analysis operations and performing residual processing with the original feature tensor to extract periodic feature components related to faults. Additionally, a constraint function for correlation feature extraction is designed to assist deep learning networks in capturing periodic features and automatically completing the description of fault knowledge. Experimental results demonstrate that the proposed method can accurately extract periodic components from bearing vibration sensor signals, automatically describe fault attributes, and play a significant role in advancing zero-shot learning tasks.
在无监督轴承故障诊断中，准确描述故障属性至关重要。然而，现有的故障属性描述依赖于人工定义，不仅劳动强度大，而且可能不够精确。因此，迫切需要开发一种自适应的自动属性描述系统。为解决这一问题，本文提出了一种基于领域知识的智能轴承故障属性描述方法，即采用深度相关特征提取网络，用于无监督故障诊断任务中属性的自动定义。通过关联分析操作提取深度特征中的周期性成分，并对原始特征张量进行残差处理，以获取与故障相关的周期性特征分量。此外，还设计了用于关联特征提取的约束函数，帮助深度学习网络捕捉周期性特征，并自动完成故障知识的描述。实验结果表明，所提出的方法能够准确提取轴承振动传感器信号中的周期性成分，自动描述故障特征，并在推进零样本学习任务方面发挥重要作用。 |
| 中文内容为网络AI翻译，专业术语未必准确，仅供参考。 |
京公网安备 11010202009788