Scholarlink

<p>Additive manufacturing is now a core component of smart manufacturing practice. However, challenges remain in developing robust and tractable design and control models underpinning the technology. In the present study, as a proof of concept, we aim for an automated method of detecting and self-rectifying the ‘stringing’ defect that can occur during Fused Filament Fabrication (FFF). This has been realized using a hybrid machine learning model, implemented by combining a convolutional neural network (CNN) and an Extreme Gradient Boosting (XGB) decision tree classifier. First, the CNN is trained offline using 5940 camera images of the stringing defect under various predefined process conditions, and during layer-by-layer material deposition, while mapping the data onto classes of print quality (low/medium/high). Then, using the output of the CNN, and based on the process control variables (retraction rate, retraction distance and nozzle temperature), the XGB classifier identifies an optimal G-code (from a predefined library) that can yield low/no stringing in the subsequent printing layer, if the defect is detected by the CNN model. Ultimately, AI-based control models can eliminate material wastage and time-consuming trial and errors towards mitigating defects observed in failed printed parts.</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

A Preliminary Step Towards Intelligent, Layer-by-Layer Self-Correction of Stringing Defect in Fused Filament Fabrication Using Limited Data

  • 期刊论文

  • 发表日期：

    2025年10月9日

查看全文
  • Rohith J. Krishnamurthy,
  • Bryn Crawford,
  • Abbas S. Milani

摘要

Additive manufacturing is now a core component of smart manufacturing practice. However, challenges remain in developing robust and tractable design and control models underpinning the technology. In the present study, as a proof of concept, we aim for an automated method of detecting and self-rectifying the ‘stringing’ defect that can occur during Fused Filament Fabrication (FFF). This has been realized using a hybrid machine learning model, implemented by combining a convolutional neural network (CNN) and an Extreme Gradient Boosting (XGB) decision tree classifier. First, the CNN is trained offline using 5940 camera images of the stringing defect under various predefined process conditions, and during layer-by-layer material deposition, while mapping the data onto classes of print quality (low/medium/high). Then, using the output of the CNN, and based on the process control variables (retraction rate, retraction distance and nozzle temperature), the XGB classifier identifies an optimal G-code (from a predefined library) that can yield low/no stringing in the subsequent printing layer, if the defect is detected by the CNN model. Ultimately, AI-based control models can eliminate material wastage and time-consuming trial and errors towards mitigating defects observed in failed printed parts.