Scholarlink

<p>Inspection of circuit waveform measurements is essential to ensure electronic product standards but is often labor-intensive, especially in large-scale manufacturing. This article presents an automated waveform anomaly detection methodology based on a deep learning driven converting and correction framework, specifically designed to enhance intelligent manufacturing by enabling real-time inspection control, reducing yield losses, and integrating with AI-driven production systems. Comparative experiments against four state-of-the-art methods, including FastFlow, PatchCore, VT-ADL, and Dinomaly, demonstrate that the proposed approach attains a 96.66% F1-score across multiple anomaly categories. A key innovation is the mutual transformation mechanism, designed to preserve signal integrity while preventing direct replication in correlated waveforms. By integrating a Long Short-Term Memory (LSTM) network with a U-Net architecture, the method simultaneously captures spatial and temporal dependencies, and an attention block further enhances focus on key waveform features. A correction model iteratively refines the transformed waveforms, further enhancing detection accuracy. Evaluations across various error types, such as trigger error (99.72%), overcurrent (99.77%), and color change (99.99%), highlight the robustness of this system. This framework offers an efficient solution for waveform anomaly detection, making it particularly well-suited for high-volume electronic product testing and quality assurance in intelligent manufacturing environments, supporting Industry 4.0/5.0 goals of automation and predictive maintenance.</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

Deep learning-based framework for automated circuit waveform anomaly detection in intelligent manufacturing

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月27日

查看全文
  • Hao-Ming Hsu,
  • Kai-Hsiang Chang,
  • Wei-Che Chang,
  • Chia-Hsun Lee,
  • Hung-Wei Hsueh,
  • Ching-Wei Tseng,
  • Shu-han Hsu

摘要

Inspection of circuit waveform measurements is essential to ensure electronic product standards but is often labor-intensive, especially in large-scale manufacturing. This article presents an automated waveform anomaly detection methodology based on a deep learning driven converting and correction framework, specifically designed to enhance intelligent manufacturing by enabling real-time inspection control, reducing yield losses, and integrating with AI-driven production systems. Comparative experiments against four state-of-the-art methods, including FastFlow, PatchCore, VT-ADL, and Dinomaly, demonstrate that the proposed approach attains a 96.66% F1-score across multiple anomaly categories. A key innovation is the mutual transformation mechanism, designed to preserve signal integrity while preventing direct replication in correlated waveforms. By integrating a Long Short-Term Memory (LSTM) network with a U-Net architecture, the method simultaneously captures spatial and temporal dependencies, and an attention block further enhances focus on key waveform features. A correction model iteratively refines the transformed waveforms, further enhancing detection accuracy. Evaluations across various error types, such as trigger error (99.72%), overcurrent (99.77%), and color change (99.99%), highlight the robustness of this system. This framework offers an efficient solution for waveform anomaly detection, making it particularly well-suited for high-volume electronic product testing and quality assurance in intelligent manufacturing environments, supporting Industry 4.0/5.0 goals of automation and predictive maintenance.