Scholarlink

<p>Recent advances in deep learning have demonstrated exceptional potential for intelligent fault diagnosis in mechanical systems. However, challenges such as limited sample availability, significant data distribution variations, and the reliance of existing feature extraction methods on marginal distributions pose significant barriers to the practical application of deep learning-based fault diagnosis techniques. To address these issues, we propose a Joint Distribution domain adaptive Brownian Prototype Network (JDBPN), which integrates metric learning and meta-learning for fault diagnosis in rotating machinery across multiple domains. First, we utilize the Brownian Correlation Metric Network (BCMN) with joint distribution to extract deep features, overcoming the limitations of existing methods that focus only on marginal distributions by combining joint feature functions with marginal feature relationships. Second, we introduce a High-Dimensional Metric Learner to enhance classification performance among fault prototypes. This learner evaluates the similarity between high-dimensional feature vectors and pre-established fault prototypes, improving the accuracy of fault category identification during meta-training. Finally, we demonstrate the effectiveness of JDBPN in fault diagnosis tasks, including few-shot and cross-domain few-shot scenarios, using two publicly available datasets.</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

Joint distribution domain adaptation: a novel meta-learning framework for cross-domain few-shot fault diagnosis

  • 期刊论文

  • 发表日期：

    2025年11月8日

查看全文
  • Yong Zhang,
  • Mengfan Zhang,
  • Wenzhe Liu

摘要

Recent advances in deep learning have demonstrated exceptional potential for intelligent fault diagnosis in mechanical systems. However, challenges such as limited sample availability, significant data distribution variations, and the reliance of existing feature extraction methods on marginal distributions pose significant barriers to the practical application of deep learning-based fault diagnosis techniques. To address these issues, we propose a Joint Distribution domain adaptive Brownian Prototype Network (JDBPN), which integrates metric learning and meta-learning for fault diagnosis in rotating machinery across multiple domains. First, we utilize the Brownian Correlation Metric Network (BCMN) with joint distribution to extract deep features, overcoming the limitations of existing methods that focus only on marginal distributions by combining joint feature functions with marginal feature relationships. Second, we introduce a High-Dimensional Metric Learner to enhance classification performance among fault prototypes. This learner evaluates the similarity between high-dimensional feature vectors and pre-established fault prototypes, improving the accuracy of fault category identification during meta-training. Finally, we demonstrate the effectiveness of JDBPN in fault diagnosis tasks, including few-shot and cross-domain few-shot scenarios, using two publicly available datasets.