Scholarlink

<p>Induction surface hardening is widely used to enhance the mechanical performance of high-strength steels such as AISI 4340, particularly for critical components like gear teeth. However, achieving a uniform and predictable hardened layer depth in geometrically complex gear profiles remains challenging. In this study, a three-dimensional multiphysics finite element model was developed in COMSOL Multiphysics to simulate the coupled electromagnetic and thermal behavior of the induction hardening process and was experimentally validated on AISI 4340 gear teeth, demonstrating high accuracy in predicting temperature distributions at both the tooth tip and root. Key process parameters, including excitation frequency, imposed current density, gear diameter, and tooth flank angle, were systematically varied to generate a structured simulation dataset. Statistical analysis identified excitation frequency as the dominant factor governing hardened layer depth. The resulting dataset was subsequently used to train an artificial neural network model for direct prediction of hardened depth at the tooth tip and root. The neural network achieved high predictive accuracy, with coefficients of determination exceeding 98% and mean relative errors below 12% for both locations, significantly outperforming regression-based baseline models. The proposed integration of validated numerical simulation, statistical analysis, and neural network-based prediction provides an efficient and scalable framework for hardened depth estimation, offering strong potential for process optimization and real time control in industrial induction hardening of gear components.</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

Data-driven prediction of hardened depth in 4340 steel gear teeth subjected to induction hardening using neural networks

  • 期刊论文

  • 发表日期：

    2026年3月15日

查看全文
  • Bouchra Khalij,
  • Narges Omidi,
  • Abderrazak El Ouafi,
  • Ahmed Chebak

摘要

Induction surface hardening is widely used to enhance the mechanical performance of high-strength steels such as AISI 4340, particularly for critical components like gear teeth. However, achieving a uniform and predictable hardened layer depth in geometrically complex gear profiles remains challenging. In this study, a three-dimensional multiphysics finite element model was developed in COMSOL Multiphysics to simulate the coupled electromagnetic and thermal behavior of the induction hardening process and was experimentally validated on AISI 4340 gear teeth, demonstrating high accuracy in predicting temperature distributions at both the tooth tip and root. Key process parameters, including excitation frequency, imposed current density, gear diameter, and tooth flank angle, were systematically varied to generate a structured simulation dataset. Statistical analysis identified excitation frequency as the dominant factor governing hardened layer depth. The resulting dataset was subsequently used to train an artificial neural network model for direct prediction of hardened depth at the tooth tip and root. The neural network achieved high predictive accuracy, with coefficients of determination exceeding 98% and mean relative errors below 12% for both locations, significantly outperforming regression-based baseline models. The proposed integration of validated numerical simulation, statistical analysis, and neural network-based prediction provides an efficient and scalable framework for hardened depth estimation, offering strong potential for process optimization and real time control in industrial induction hardening of gear components.