Scholarlink

<p>Accurate fault diagnosis of rolling bearings is crucial for ensuring the reliability of rotating machinery, yet its performance is often degraded by noise and fluctuating operating conditions. To address this challenge, we propose a robust diagnostic framework that integrates advanced signal preprocessing with an enhanced residual shrinkage network. First, ensemble empirical mode decomposition (EEMD) is employed to decompose vibration signals, and correlation-based selection followed by principal component analysis (PCA) ensures compact and informative feature representation. Next, an adaptive soft thresholding mechanism dynamically adjusts threshold and slope parameters for improved noise suppression. A dual-pooling attention strategy further enhances generalization by focusing on fault-relevant features. Finally, a deep residual shrinkage network with improved blocks strengthens feature learning and classification robustness. Experimental validation on the benchmark Case Western Reserve University (CWRU) dataset demonstrates that the proposed method not only outperforms existing models but also provides a reliable tool for predictive maintenance, thereby enhancing operational efficiency and reducing unplanned downtime in industrial machinery.</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

A robust deep learning framework for bearing fault diagnosis using residual shrinkage networks and signal denoising

  • 期刊论文

  • 发表日期：

    2026年3月8日

查看全文
  • Sujit Kumar,
  • Bam Bahadur Sinha

摘要

Accurate fault diagnosis of rolling bearings is crucial for ensuring the reliability of rotating machinery, yet its performance is often degraded by noise and fluctuating operating conditions. To address this challenge, we propose a robust diagnostic framework that integrates advanced signal preprocessing with an enhanced residual shrinkage network. First, ensemble empirical mode decomposition (EEMD) is employed to decompose vibration signals, and correlation-based selection followed by principal component analysis (PCA) ensures compact and informative feature representation. Next, an adaptive soft thresholding mechanism dynamically adjusts threshold and slope parameters for improved noise suppression. A dual-pooling attention strategy further enhances generalization by focusing on fault-relevant features. Finally, a deep residual shrinkage network with improved blocks strengthens feature learning and classification robustness. Experimental validation on the benchmark Case Western Reserve University (CWRU) dataset demonstrates that the proposed method not only outperforms existing models but also provides a reliable tool for predictive maintenance, thereby enhancing operational efficiency and reducing unplanned downtime in industrial machinery.