Scholarlink

<p>Additive Manufacturing (AM), particularly Fused Deposition Modeling (FDM/FFF), is widely adopted for rapid prototyping and cost-effective part production, yet its broader and more repeatable use is still constrained by frequent print defects, operator-dependent troubleshooting, and limited in-situ quality assurance, especially on desktop and entry-level 3D printers where built-in monitoring is typically unavailable. In the context of Industry 4.0 and smart manufacturing, this motivates low-cost, Machine learning (ML) and Artificial intelligence (AI) -enabled metrology approaches that can generate actionable process information and support more standardized, data-driven printing workflows. This study presents a real-time, in-situ defect detection framework specifically designed for low-cost extrusion printers, integrating a Creality Ender 5 Pro 3D printer with an overhead optical camera and lightweight computational modules for continuous layer-wise image processing. A machine learning image-classification model is trained to detect five printing states: good prints, high bed level, low bed level, high flow rate, and low flow rate, enabling early anomaly identification and timely intervention to reduce failed builds and filament waste. The results further highlight that model reliability is strongly governed by dataset structure, consistent labeling, and controlled parameter definition, reinforcing data discipline as a prerequisite for dependable low-cost AI monitoring. In addition, material consumption and energy use are evaluated to quantify the operational value of early defect detection, illustrating how low-cost predictive analytics can reduce waste and improve process efficiency. The current prototype uses a single top-down RGB view and is therefore most effective for early-layer, surface-visible signatures; later-layer occlusion and subsurface/interior defects will require multi-view and/or multi-modal sensing in future work. Overall, the proposed system demonstrates a practical pathway for bringing accessible, scalable in-situ monitoring to the desktop FDM ecosystem and for generating traceable quality evidence that supports more repeatable and qualification-oriented AM practice.</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

Machine learning-enabled in-situ defect detection in extrusion-based additive manufacturing: a step toward qualification and certification

  • 期刊论文

  • 发表日期：

    2026年3月13日

查看全文
  • Md Junaid Shekh,
  • Cooper Parson,
  • Anubhav Mishra,
  • Subrata Sanyal,
  • Hitesh D. Vora

摘要

Additive Manufacturing (AM), particularly Fused Deposition Modeling (FDM/FFF), is widely adopted for rapid prototyping and cost-effective part production, yet its broader and more repeatable use is still constrained by frequent print defects, operator-dependent troubleshooting, and limited in-situ quality assurance, especially on desktop and entry-level 3D printers where built-in monitoring is typically unavailable. In the context of Industry 4.0 and smart manufacturing, this motivates low-cost, Machine learning (ML) and Artificial intelligence (AI) -enabled metrology approaches that can generate actionable process information and support more standardized, data-driven printing workflows. This study presents a real-time, in-situ defect detection framework specifically designed for low-cost extrusion printers, integrating a Creality Ender 5 Pro 3D printer with an overhead optical camera and lightweight computational modules for continuous layer-wise image processing. A machine learning image-classification model is trained to detect five printing states: good prints, high bed level, low bed level, high flow rate, and low flow rate, enabling early anomaly identification and timely intervention to reduce failed builds and filament waste. The results further highlight that model reliability is strongly governed by dataset structure, consistent labeling, and controlled parameter definition, reinforcing data discipline as a prerequisite for dependable low-cost AI monitoring. In addition, material consumption and energy use are evaluated to quantify the operational value of early defect detection, illustrating how low-cost predictive analytics can reduce waste and improve process efficiency. The current prototype uses a single top-down RGB view and is therefore most effective for early-layer, surface-visible signatures; later-layer occlusion and subsurface/interior defects will require multi-view and/or multi-modal sensing in future work. Overall, the proposed system demonstrates a practical pathway for bringing accessible, scalable in-situ monitoring to the desktop FDM ecosystem and for generating traceable quality evidence that supports more repeatable and qualification-oriented AM practice.