Scholarlink

<p>In advanced manufacturing, tool wear monitoring plays a pivotal role in sustaining process reliability, reducing downtime, and preventing unexpected failures, challenges that become more critical when machining difficult-to-cut materials such as Ti6Al4V, widely employed in aerospace components. This study explores the prediction of tool wear using machine learning classifiers, with an emphasis on enhancing the transparency of predictive outcomes through interpretability techniques. Controlled experiments were conducted by varying cutting speed (vc [m/min]), feed rate (<i>f</i> [mm/rev]), and cutting force (Fc [N]), parameters closely linked to tool degradation in dry machining conditions. The models assessed included Random Forest, LightGBM, Support Vector Machine (SVM), and Logistic Regression. For interpretability, LIME was applied to generate local explanations of individual predictions, while SHAP provided both local and global feature importance analysis, enabling a broader understanding of model behavior. The analysis revealed that Fc [N] and <i>f</i> [mm/rev] were the most influential features. LightGBM yielded the highest classification accuracy, with Random Forest performing similarly. Notably, the exclusion of vc [m/min] and <i>f</i> [mm/rev] led to performance decline, underlining their predictive value. By integrating AI with domain knowledge, this work contributes to more transparent and trustworthy decision-making in machining, an essential step toward intelligent manufacturing systems capable of handling complex material behaviors.</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

Wear classification in Ti6Al4V machining using explainable machine learning: a decision support approach

  • 期刊论文

  • 发表日期：

    2026年1月6日

查看全文
  • Alex Fernandes de Souza,
  • Leonardo Lourenço de Souza,
  • Juliana Helena Daroz Gaudêncio,
  • Carlos Eduardo Sanches da Silva,
  • Filipe Alves Neto Verri,
  • Pedro Paulo Balestrassi

摘要

In advanced manufacturing, tool wear monitoring plays a pivotal role in sustaining process reliability, reducing downtime, and preventing unexpected failures, challenges that become more critical when machining difficult-to-cut materials such as Ti6Al4V, widely employed in aerospace components. This study explores the prediction of tool wear using machine learning classifiers, with an emphasis on enhancing the transparency of predictive outcomes through interpretability techniques. Controlled experiments were conducted by varying cutting speed (vc [m/min]), feed rate (f [mm/rev]), and cutting force (Fc [N]), parameters closely linked to tool degradation in dry machining conditions. The models assessed included Random Forest, LightGBM, Support Vector Machine (SVM), and Logistic Regression. For interpretability, LIME was applied to generate local explanations of individual predictions, while SHAP provided both local and global feature importance analysis, enabling a broader understanding of model behavior. The analysis revealed that Fc [N] and f [mm/rev] were the most influential features. LightGBM yielded the highest classification accuracy, with Random Forest performing similarly. Notably, the exclusion of vc [m/min] and f [mm/rev] led to performance decline, underlining their predictive value. By integrating AI with domain knowledge, this work contributes to more transparent and trustworthy decision-making in machining, an essential step toward intelligent manufacturing systems capable of handling complex material behaviors.