Scholarlink

<p>Fused Deposition Modelling, a key additive manufacturing process, faces persistent challenges in real-time defect detection, particularly for defects such as stringing and spaghetti-like defects. This study presents an approach that integrates transformer-based models, Vision Transformer, Swin Transformer, and Data-efficient Image Transformer, with traditional deep learning architectures such as CNN, InceptionNet, and UNet. A hybrid framework, combining these models with Contrast Limited Adaptive Histogram Equalisation (CLAHE) pre-processing, enhances defect visibility and detection accuracy. Experimental results have shown that DeiT with CLAHE has achieved an average accuracy of 97.6%±0.7, along with high sensitivity and precision. The findings highlight that transformer-based methods can significantly improve defect detection reliability, reduce material waste, and support automation in FDM quality assurance.</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

Real-Time Defect Detection in Fused Deposition 3D Printing Process Using Transformer-based Models

  • 期刊论文

  • 发表日期：

    2026年1月21日

查看全文
  • Aarush Aggarwal,
  • Shaurya Bhardwaj,
  • Vinay Arora,
  • Jaswinder Singh Saini,
  • Sachin Singh

摘要

Fused Deposition Modelling, a key additive manufacturing process, faces persistent challenges in real-time defect detection, particularly for defects such as stringing and spaghetti-like defects. This study presents an approach that integrates transformer-based models, Vision Transformer, Swin Transformer, and Data-efficient Image Transformer, with traditional deep learning architectures such as CNN, InceptionNet, and UNet. A hybrid framework, combining these models with Contrast Limited Adaptive Histogram Equalisation (CLAHE) pre-processing, enhances defect visibility and detection accuracy. Experimental results have shown that DeiT with CLAHE has achieved an average accuracy of 97.6%±0.7, along with high sensitivity and precision. The findings highlight that transformer-based methods can significantly improve defect detection reliability, reduce material waste, and support automation in FDM quality assurance.