Scholarlink

<p>Machining is a key process in the manufacturing industry, significantly influencing production cost and efficiency. With the rise of Industry 4.0, artificial intelligence (AI) and data-driven methods have enabled advances in process optimization and tool design. However, accurately modeling machining processes remains a challenge due to their nonlinear and multivariate nature. This study proposes an integrated framework combining the Finite Element Method (FEM), machine learning (ML), and optimization algorithms to minimize cutting forces and support intelligent cutting-tool design. FEM simulations of orthogonal cutting were performed by varying rake angle, cutting speed, and Johnson–Cook material parameters to generate a numerical dataset. Although constrained by the computational cost inherent to finite element analysis, the cutting forces were obtained using a well-established thermo-mechanically coupled model, which accounts for strain-rate and temperature-dependent Johnson–Cook plasticity and a physically based, experimentally calibrated damage law. Surrogate ML models—tree-based and regression—were trained to predict cutting forces, achieving <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\({R}^{2}\)</EquationSource> </InlineEquation> &gt; 0.95. These models were then coupled with Powell’s method and Genetic Algorithms to iteratively determine optimal tool geometries, feeding the optimized results back into the simulation process to verify performance. The Lasso model maintained prediction errors below 10% for unseen configurations, confirming its robustness and generalization ability. The proposed FEM–ML–optimization loop goes beyond conventional predictive approaches by enabling adaptive design exploration without rerunning all FEM simulations, thereby reducing computational cost. This framework establishes a foundation for data-efficient and scalable tool-design methodologies and opens opportunities for future research using alternative optimizers such as Mesh Adaptive Direct Search (MADS) and Bayesian Optimization.</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

Artificial intelligence-based cutting tool geometry optimization for cutting force minimization in machining

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月11日

查看全文
  • Igor Henrique de Freitas,
  • Modesto Valci Moreira Lopes,
  • Angel Biskupovic,
  • Izabel Fernanda Machado

摘要

Machining is a key process in the manufacturing industry, significantly influencing production cost and efficiency. With the rise of Industry 4.0, artificial intelligence (AI) and data-driven methods have enabled advances in process optimization and tool design. However, accurately modeling machining processes remains a challenge due to their nonlinear and multivariate nature. This study proposes an integrated framework combining the Finite Element Method (FEM), machine learning (ML), and optimization algorithms to minimize cutting forces and support intelligent cutting-tool design. FEM simulations of orthogonal cutting were performed by varying rake angle, cutting speed, and Johnson–Cook material parameters to generate a numerical dataset. Although constrained by the computational cost inherent to finite element analysis, the cutting forces were obtained using a well-established thermo-mechanically coupled model, which accounts for strain-rate and temperature-dependent Johnson–Cook plasticity and a physically based, experimentally calibrated damage law. Surrogate ML models—tree-based and regression—were trained to predict cutting forces, achieving \({R}^{2}\) > 0.95. These models were then coupled with Powell’s method and Genetic Algorithms to iteratively determine optimal tool geometries, feeding the optimized results back into the simulation process to verify performance. The Lasso model maintained prediction errors below 10% for unseen configurations, confirming its robustness and generalization ability. The proposed FEM–ML–optimization loop goes beyond conventional predictive approaches by enabling adaptive design exploration without rerunning all FEM simulations, thereby reducing computational cost. This framework establishes a foundation for data-efficient and scalable tool-design methodologies and opens opportunities for future research using alternative optimizers such as Mesh Adaptive Direct Search (MADS) and Bayesian Optimization.