Scholarlink

In this work, we share our learnings on predicting CNC manufacturing time given CAD models from a real-world dataset using machine learning models. To minimize prediction cost, we focus on extracting relevant information solely from the CAD model, eliminating the need for manual feature labeling. Our experiments reveal that a combination of hand-crafted features with those automatically extracted by the graph convolutional network UV-Net yields the most accurate predictions. Notably, our model exhibits robust performance when applied to a newer, higher-quality dataset, achieving a significant improvement of 32% in mean absolute error compared to a rule-based approach despite the challenges posed by temporal and data quality shifts.

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

Prediction of CNC Manufacturing Time Under Real-World Conditions Using Graph Convolutional Networks

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Fabio Lischka,
  • Andreas Schwarz,
  • Dominik Wiesner,
  • Christoph Wald,
  • Frank Schirmeier,
  • Ulrich Göhner

摘要

In this work, we share our learnings on predicting CNC manufacturing time given CAD models from a real-world dataset using machine learning models. To minimize prediction cost, we focus on extracting relevant information solely from the CAD model, eliminating the need for manual feature labeling. Our experiments reveal that a combination of hand-crafted features with those automatically extracted by the graph convolutional network UV-Net yields the most accurate predictions. Notably, our model exhibits robust performance when applied to a newer, higher-quality dataset, achieving a significant improvement of 32% in mean absolute error compared to a rule-based approach despite the challenges posed by temporal and data quality shifts.