Scholarlink

<p>In recent years, deep learning has made significant progress in the field of mechanical fault diagnosis, yet two critical challenges remain in practical industrial applications: the high cost of labeling fault data renders large amounts of unlabeled data difficult to utilize, and the imbalanced distribution of fault types severely impacts model performance. To address these issues, this study proposes a two-stage self-supervised learning framework (SMDN). Based on an adaptive wavelet convolutional residual network, the study constructs a self-supervised framework that integrates data augmentation and contrastive learning, effectively extracting robust features from unlabeled data. Subsequently, transfer learning is employed to transfer the pre-trained knowledge to downstream classification tasks, while a class-weighted loss function is adopted to mitigate data imbalance issues. Experimental results demonstrate that the proposed method performs exceptionally well on both the Politecnico di Torino bearing dataset and the Jilin University high-speed motor dataset. Under varying labeling rates and different degrees of imbalance, its fault diagnosis accuracy consistently outperforms existing methods, fully validating the practicality and generalizability of this approach in the field of industrial fault diagnosis.</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

Intelligent diagnosis of high-speed motors under data imbalance scenarios:self-supervised feature extraction and classification optimization

  • 期刊论文

  • 发表日期：

    2025年9月9日

查看全文
  • Fenghao Sun,
  • Guofa Li,
  • Jialong He,
  • Wei Zhang

摘要

In recent years, deep learning has made significant progress in the field of mechanical fault diagnosis, yet two critical challenges remain in practical industrial applications: the high cost of labeling fault data renders large amounts of unlabeled data difficult to utilize, and the imbalanced distribution of fault types severely impacts model performance. To address these issues, this study proposes a two-stage self-supervised learning framework (SMDN). Based on an adaptive wavelet convolutional residual network, the study constructs a self-supervised framework that integrates data augmentation and contrastive learning, effectively extracting robust features from unlabeled data. Subsequently, transfer learning is employed to transfer the pre-trained knowledge to downstream classification tasks, while a class-weighted loss function is adopted to mitigate data imbalance issues. Experimental results demonstrate that the proposed method performs exceptionally well on both the Politecnico di Torino bearing dataset and the Jilin University high-speed motor dataset. Under varying labeling rates and different degrees of imbalance, its fault diagnosis accuracy consistently outperforms existing methods, fully validating the practicality and generalizability of this approach in the field of industrial fault diagnosis.