Scholarlink

<p>Angular contact ball bearings (ACBBs) are widely used for their structural and economic advantages. However, achieving optimal design to meet performance requirements remains challenging with traditional optimal design methods, which rely on initial population selection and struggle to adapt to changing conditions. To address these challenges, this study employs deep reinforcement learning (DRL) for ACBB design. The proposed DRL framework is further enhanced by integrating transfer learning (TL), which accelerates learning when transitioning to new conditions by leveraging knowledge from previously trained models. The framework was evaluated for its applicability, sensitivity to initial states, and adaptability to varying loads. The DRL agent identified lightweight ACBBs satisfying life and contact stress requirements, closely matching true solutions. Furthermore, TL enabled rapid adaptation to new load conditions, highlighting the efficiency of the proposed approach. These findings suggest that DRL and TL provide a flexible and reliable methodology for ACBB design in dynamic environments.</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

Deep reinforcement learning driven design optimization of angular contact ball bearings

  • 期刊论文

  • 发表日期：

    2025年5月7日

查看全文
  • Beom-Soo Kim,
  • Doyeop Kwon,
  • Dongu Im,
  • Jung-Ho Park,
  • Young-Jun Park

摘要

Angular contact ball bearings (ACBBs) are widely used for their structural and economic advantages. However, achieving optimal design to meet performance requirements remains challenging with traditional optimal design methods, which rely on initial population selection and struggle to adapt to changing conditions. To address these challenges, this study employs deep reinforcement learning (DRL) for ACBB design. The proposed DRL framework is further enhanced by integrating transfer learning (TL), which accelerates learning when transitioning to new conditions by leveraging knowledge from previously trained models. The framework was evaluated for its applicability, sensitivity to initial states, and adaptability to varying loads. The DRL agent identified lightweight ACBBs satisfying life and contact stress requirements, closely matching true solutions. Furthermore, TL enabled rapid adaptation to new load conditions, highlighting the efficiency of the proposed approach. These findings suggest that DRL and TL provide a flexible and reliable methodology for ACBB design in dynamic environments.