Scholarlink

<p>In machining operations, tool condition monitoring must remain simple and accurate despite changes in jobs, materials, and setups; however, most existing data-driven models are either overly complex or degrade under variability across machines, materials, cutting parameters, and sensor configurations, limiting industrial deployment potential without frequent retraining. To address this, this paper focuses on domain generalization for real-time tool wear monitoring under data imbalance and proposes Conditional-Generative Augmentation for Generalization (C-GAG), a conditional GAN-based framework that synthesizes physically consistent sensor–process samples conditioned on tool wear, depth of cut, feed rate, and workpiece material. By augmenting small, imbalanced run-to-failure datasets, C-GAG enables more generalizable classification, particularly under unseen machining conditions. Validation on cast iron and J45 steel demonstrates substantial improvements: for cast iron, F1 increases from 44% to 66%, accuracy from 57% to 73%, and AUC reaches 97%, while for J45 steel, F1 improves from 4% to 47%, accuracy from 10% to 75%, and AUC from 66% to 92%. A sensor-economy analysis further shows that competitive accuracy can be maintained with reduced sensor configurations—66.7% for cast iron and 76.7% for J45, lowering deployment cost. Overall, C-GAG improves robustness to process variability without requiring complex sensing, additional labeling, or extensive re-tuning, advancing domain-general tool wear monitoring toward practical, industrial deployment potential.</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

Conditional-generative augmentation-based generalization (C-GAG): towards generalizable wear monitoring in machining operations

  • 期刊论文

  • 发表日期：

    2026年7月3日

查看全文
  • Harmandeep Singh,
  • Saeed Moghadasi,
  • Mohamed Abubakr Hassan,
  • Chi-Guhn Lee,
  • Hussien Hegab

摘要

In machining operations, tool condition monitoring must remain simple and accurate despite changes in jobs, materials, and setups; however, most existing data-driven models are either overly complex or degrade under variability across machines, materials, cutting parameters, and sensor configurations, limiting industrial deployment potential without frequent retraining. To address this, this paper focuses on domain generalization for real-time tool wear monitoring under data imbalance and proposes Conditional-Generative Augmentation for Generalization (C-GAG), a conditional GAN-based framework that synthesizes physically consistent sensor–process samples conditioned on tool wear, depth of cut, feed rate, and workpiece material. By augmenting small, imbalanced run-to-failure datasets, C-GAG enables more generalizable classification, particularly under unseen machining conditions. Validation on cast iron and J45 steel demonstrates substantial improvements: for cast iron, F1 increases from 44% to 66%, accuracy from 57% to 73%, and AUC reaches 97%, while for J45 steel, F1 improves from 4% to 47%, accuracy from 10% to 75%, and AUC from 66% to 92%. A sensor-economy analysis further shows that competitive accuracy can be maintained with reduced sensor configurations—66.7% for cast iron and 76.7% for J45, lowering deployment cost. Overall, C-GAG improves robustness to process variability without requiring complex sensing, additional labeling, or extensive re-tuning, advancing domain-general tool wear monitoring toward practical, industrial deployment potential.