Scholarlink

<p>Powder Bed Fusion Laser Beam (PBF-LB/M) is a key additive manufacturing process for producing high-performance metal parts. However, the formation of porosity defects during fabrication remains a major challenge, compromising structural integrity and fatigue life. Although in situ monitoring systems leveraging optical, acoustic, and thermal modalities have advanced, they often struggle to detect the phenomena leading to porosity with sufficient temporal resolution due to the intricate dynamics of the melt pool. Furthermore, conventional machine learning methods typically depend on densely labeled datasets, which are impractical to obtain in real-time industrial environments. Current in situ monitoring solutions, though increasingly sophisticated, face challenges in detecting such defects at fine temporal scales. To address these limitations, this study presents a novel defect detection framework using Outside-In Multi-Instance Learning (OI-MIL) with Acoustic Ultrasonic (AU) and photodiode sensors, enabling coarse detection at 50&#xa0;ms intervals and fine-grained porosity inference every 5&#xa0;ms. The AU sensor shows particularly strong performance, although sensor fusion does not improve accuracy as expected. These results highlight the effectiveness of AU sensing for defect detection and offer valuable insights into signal redundancy and practical sensor deployment strategies for industrial applications.</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

Porosity detection on PBF-LB/M using “Outside-in” multi-instance learning based on ultrasonic acoustic and photodiode sensors

  • 期刊论文

  • 发表日期：

    2025年11月12日

查看全文
  • El Arbi Hajjioui,
  • Foued Abroug,
  • Maher Baili,
  • Lionel Arnaud,
  • Khanh Nguyen,
  • Kamal Medjaher

摘要

Powder Bed Fusion Laser Beam (PBF-LB/M) is a key additive manufacturing process for producing high-performance metal parts. However, the formation of porosity defects during fabrication remains a major challenge, compromising structural integrity and fatigue life. Although in situ monitoring systems leveraging optical, acoustic, and thermal modalities have advanced, they often struggle to detect the phenomena leading to porosity with sufficient temporal resolution due to the intricate dynamics of the melt pool. Furthermore, conventional machine learning methods typically depend on densely labeled datasets, which are impractical to obtain in real-time industrial environments. Current in situ monitoring solutions, though increasingly sophisticated, face challenges in detecting such defects at fine temporal scales. To address these limitations, this study presents a novel defect detection framework using Outside-In Multi-Instance Learning (OI-MIL) with Acoustic Ultrasonic (AU) and photodiode sensors, enabling coarse detection at 50 ms intervals and fine-grained porosity inference every 5 ms. The AU sensor shows particularly strong performance, although sensor fusion does not improve accuracy as expected. These results highlight the effectiveness of AU sensing for defect detection and offer valuable insights into signal redundancy and practical sensor deployment strategies for industrial applications.