Scholarlink

<p>In modern industrial environments, efficient and non-invasive monitoring of machinery operations is crucial for ensuring process optimization and early fault detection. Traditional physical sensors, while effective, can be costly and impractical to deploy extensively across complex systems. This paper introduces an innovative approach leveraging non-negative matrix factorization (NMF) to create acoustic virtual sensors that analyze sound spectrograms for real-time industrial process monitoring. By decomposing acoustic signals captured from machinery into distinct spectral components, the proposed method enables the detection of specific operational phases and potential anomalies. While the methodology is demonstrated using a plastic injection molding machine, it is designed to be adaptable to a wide range of industrial processes where machinery generates distinct acoustic signatures. The approach involves capturing high-fidelity acoustic data, applying NMF to extract activation matrices that represent unique acoustic patterns, and using clustering techniques to ensure robust identification of operational states across different environments. This generalizable framework allows for scalable monitoring solutions across various industrial applications, from manufacturing lines to heavy machinery operations. This study highlights the potential of acoustic virtual sensors as a cost-effective, scalable solution for industrial monitoring, offering new possibilities for predictive maintenance and anomaly detection in diverse manufacturing environments.</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

Acoustic virtual sensors for industrial process monitoring using non-negative matrix factorization

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年8月1日

查看全文
  • Clara Luzon-Alvarez,
  • Maximo Cobos,
  • Jesus Lopez-Ballester,
  • Ana M. Torres-Aranda,
  • Francesc J. Ferri

摘要

In modern industrial environments, efficient and non-invasive monitoring of machinery operations is crucial for ensuring process optimization and early fault detection. Traditional physical sensors, while effective, can be costly and impractical to deploy extensively across complex systems. This paper introduces an innovative approach leveraging non-negative matrix factorization (NMF) to create acoustic virtual sensors that analyze sound spectrograms for real-time industrial process monitoring. By decomposing acoustic signals captured from machinery into distinct spectral components, the proposed method enables the detection of specific operational phases and potential anomalies. While the methodology is demonstrated using a plastic injection molding machine, it is designed to be adaptable to a wide range of industrial processes where machinery generates distinct acoustic signatures. The approach involves capturing high-fidelity acoustic data, applying NMF to extract activation matrices that represent unique acoustic patterns, and using clustering techniques to ensure robust identification of operational states across different environments. This generalizable framework allows for scalable monitoring solutions across various industrial applications, from manufacturing lines to heavy machinery operations. This study highlights the potential of acoustic virtual sensors as a cost-effective, scalable solution for industrial monitoring, offering new possibilities for predictive maintenance and anomaly detection in diverse manufacturing environments.