Scholarlink

<p>Analog circuit soft fault diagnosis remains crucial for system reliability but is hindered by weak fault signatures and significant noise interference. This study introduces a novel framework that converts one-dimensional circuit signals into two-dimensional Gramian Angular Field (GAF) images, recasting the diagnosis as an image classification task. The core of the method is a Multi-scale Deep Residual Shrinkage Network (MS-DARSN), which leverages an integrated dual-attention mechanism to adaptively suppress noise and refine discriminative features. Evaluated on a four-op-amp high-pass filter and a Sallen-Key band-pass filter, the proposed GAF-MS-DARSN model achieved high accuracies of 99.87% and 99.81%, respectively, surpassing benchmark methods in precision and F1-score. Robustness evaluation via five-fold cross-validation under 10 dB noise conditions demonstrates that the model maintains an accuracy of 95.47%, reflecting its strong generalization capability and practical potential.</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

Analog Circuit Fault Diagnosis Based on Gramian Angular Field and Multi-Scale Dual Attention Residual Shrinkage Network

  • 期刊论文

  • 发表日期：

    2026年2月9日

查看全文
  • Gefei Duan,
  • Yajing Gao,
  • Rui Yang,
  • Junying Yun,
  • Meili Rao,
  • Dong Ye,
  • Xinyi Wei

摘要

Analog circuit soft fault diagnosis remains crucial for system reliability but is hindered by weak fault signatures and significant noise interference. This study introduces a novel framework that converts one-dimensional circuit signals into two-dimensional Gramian Angular Field (GAF) images, recasting the diagnosis as an image classification task. The core of the method is a Multi-scale Deep Residual Shrinkage Network (MS-DARSN), which leverages an integrated dual-attention mechanism to adaptively suppress noise and refine discriminative features. Evaluated on a four-op-amp high-pass filter and a Sallen-Key band-pass filter, the proposed GAF-MS-DARSN model achieved high accuracies of 99.87% and 99.81%, respectively, surpassing benchmark methods in precision and F1-score. Robustness evaluation via five-fold cross-validation under 10 dB noise conditions demonstrates that the model maintains an accuracy of 95.47%, reflecting its strong generalization capability and practical potential.