Scholarlink

<p>In the context of Industry 4.0, enabling technologies such as Additive Manufacturing (AM) and Digital Twins (DT) have been transforming traditional manufacturing processes through AI-driven data analysis and real-time decision-making. This paper presents a novel defect classification method for metal-deposited parts in a robotic cell using Laser Metal Deposition with wire (LMD-wire). Unlike previous approaches, this work introduces a structured classification framework that categorizes defects into five specific classes (Balling, Dripping, Necking, Overbuilding, and Stubbing) and four quality levels, enabling automated quality control and process optimization. The proposed system is based on two Convolutional Neural Network (CNN) architectures: YOLOv5s (You Only Look Once) and Faster R-CNN (Faster Region-based Convolutional Neural Network), which are trained to detect deposition defects in LMD-wire printed parts. A comparative analysis of these models is conducted to evaluate their performance. An average defect detection accuracy of 0.932 was achieved with YOLOv5s, compared to 0.872 with Faster R-CNN. Defects such as Overbuilding were detected with higher accuracy using Faster R-CNN, while the Necking defect was more accurately identified by YOLOv5s. The experiments and results suggest the combined use of both models for better prediction of surface defects in parts printed using LMD-wire. In addition, the system incorporates a prescriptive mechanism that suggests corrective actions based on detected defects, improves defect diagnosis, and contributes to improved manufacturing reliability.</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

Automated defect classification in additive manufacturing LMD-wire using deep learning

  • 期刊论文

  • 发表日期：

    2025年7月2日

查看全文
  • Alberto José Álvares,
  • Brayan Stiven Figueroa Betancourth,
  • João Vítor Arantes Cabral,
  • Igor Lacroix

摘要

In the context of Industry 4.0, enabling technologies such as Additive Manufacturing (AM) and Digital Twins (DT) have been transforming traditional manufacturing processes through AI-driven data analysis and real-time decision-making. This paper presents a novel defect classification method for metal-deposited parts in a robotic cell using Laser Metal Deposition with wire (LMD-wire). Unlike previous approaches, this work introduces a structured classification framework that categorizes defects into five specific classes (Balling, Dripping, Necking, Overbuilding, and Stubbing) and four quality levels, enabling automated quality control and process optimization. The proposed system is based on two Convolutional Neural Network (CNN) architectures: YOLOv5s (You Only Look Once) and Faster R-CNN (Faster Region-based Convolutional Neural Network), which are trained to detect deposition defects in LMD-wire printed parts. A comparative analysis of these models is conducted to evaluate their performance. An average defect detection accuracy of 0.932 was achieved with YOLOv5s, compared to 0.872 with Faster R-CNN. Defects such as Overbuilding were detected with higher accuracy using Faster R-CNN, while the Necking defect was more accurately identified by YOLOv5s. The experiments and results suggest the combined use of both models for better prediction of surface defects in parts printed using LMD-wire. In addition, the system incorporates a prescriptive mechanism that suggests corrective actions based on detected defects, improves defect diagnosis, and contributes to improved manufacturing reliability.