Scholarlink

<p>With the rapid development of deep learning, a variety of intelligent techniques have been successfully applied to fault diagnosis, yielding promising results. However, most existing methods rely on data collected under ideal conditions, which limits their applicability in real-world industrial settings characterized by limited data availability, difficulties in labeling, and significant operating condition discrepancies. To address these challenges, this paper proposes a Domain-Adversarial Semi-Supervised Prototypical Network (DA-SSPN). The proposed approach integrates the prototypical network architecture with a domain-adversarial training strategy, and employs a synergistic optimization of a feature extractor, a label classifier, and a domain discriminator to effectively extract domain-invariant features. During prototype construction, unlabeled samples are incorporated into the learning of class centers, and a shifting term is introduced to mitigate the distributional bias between the support set and the query set. This enhances classification stability and diagnostic accuracy across varying operating conditions. The proposed method is evaluated on several cross-domain tasks and compared with existing approaches. Experimental results demonstrate that DA-SSPN achieves superior classification performance and generalization capability.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Semi-supervised prototype network with domain adversarial for few-shot fault diagnosis

  • 期刊论文

  • 发表日期：

    2025年7月28日

查看全文
  • Haohao Song,
  • Jie Yang,
  • Xiaowei Wan,
  • Anke Xue

摘要

With the rapid development of deep learning, a variety of intelligent techniques have been successfully applied to fault diagnosis, yielding promising results. However, most existing methods rely on data collected under ideal conditions, which limits their applicability in real-world industrial settings characterized by limited data availability, difficulties in labeling, and significant operating condition discrepancies. To address these challenges, this paper proposes a Domain-Adversarial Semi-Supervised Prototypical Network (DA-SSPN). The proposed approach integrates the prototypical network architecture with a domain-adversarial training strategy, and employs a synergistic optimization of a feature extractor, a label classifier, and a domain discriminator to effectively extract domain-invariant features. During prototype construction, unlabeled samples are incorporated into the learning of class centers, and a shifting term is introduced to mitigate the distributional bias between the support set and the query set. This enhances classification stability and diagnostic accuracy across varying operating conditions. The proposed method is evaluated on several cross-domain tasks and compared with existing approaches. Experimental results demonstrate that DA-SSPN achieves superior classification performance and generalization capability.