Scholarlink

<p>Titanium/aluminum (Ti/Al) laminated composites are attractive for lightweight, high-performance structures, yet optimizing ultrasonic additive manufacturing (UAM) remains challenging because interfacial bonding depends on strongly coupled process variables and dissimilar-metal interactions, while experimental datasets are often limited. Here, Ti/Al laminates are fabricated by UAM, and a GPR-based surrogate modeling workflow is used to guide process-window identification under data-scarce conditions. Gaussian process regression (GPR) is trained on a small experimental dataset and evaluated by leave-one-out cross-validation, providing both mean predictions and uncertainty estimates for Bayesian optimization; a neural-network model is included only as a deterministic baseline. The model-guided search identifies an operating region centered at 10&#xa0;mm/s travel speed, 1500&#xa0;N normal force, and ~ 40&#xa0;μm amplitude, which is confirmed experimentally with a peak peel strength of ~ 145&#xa0;N, and high bonding is maintained within a narrow amplitude neighborhood (~ 38–42&#xa0;μm). Importantly, interfacial peel strength alone does not predict bulk toughness: samples welded at higher amplitude (e.g., 42&#xa0;μm) retain high peel strength, but exhibit reduced bending energy absorption, consistent with excessive work hardening in the Al layers. EBSD and electron microscopy reveal coupled grain-scale gradients within the Al layers and along the build direction (fine–coarse–fine across-layer variation and progressive refinement through thickness), consistent with cyclic ultrasonic deformation and continuous dynamic recrystallization. Overall, the study links UAM parameters, interfacial bonding, bulk response, and gradient microstructure evolution and demonstrates a practical, data-efficient route for process exploration of dissimilar-metal laminates.</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

Machine learning-aided process optimization and gradient microstructure evolution in ultrasonic additive manufacturing of Ti/Al layered composites

  • 期刊论文

  • 发表日期：

    2026年7月25日

查看全文
  • Qiu Wang,
  • Xianheng Deng,
  • Fei Yin,
  • Lin Hua

摘要

Titanium/aluminum (Ti/Al) laminated composites are attractive for lightweight, high-performance structures, yet optimizing ultrasonic additive manufacturing (UAM) remains challenging because interfacial bonding depends on strongly coupled process variables and dissimilar-metal interactions, while experimental datasets are often limited. Here, Ti/Al laminates are fabricated by UAM, and a GPR-based surrogate modeling workflow is used to guide process-window identification under data-scarce conditions. Gaussian process regression (GPR) is trained on a small experimental dataset and evaluated by leave-one-out cross-validation, providing both mean predictions and uncertainty estimates for Bayesian optimization; a neural-network model is included only as a deterministic baseline. The model-guided search identifies an operating region centered at 10 mm/s travel speed, 1500 N normal force, and ~ 40 μm amplitude, which is confirmed experimentally with a peak peel strength of ~ 145 N, and high bonding is maintained within a narrow amplitude neighborhood (~ 38–42 μm). Importantly, interfacial peel strength alone does not predict bulk toughness: samples welded at higher amplitude (e.g., 42 μm) retain high peel strength, but exhibit reduced bending energy absorption, consistent with excessive work hardening in the Al layers. EBSD and electron microscopy reveal coupled grain-scale gradients within the Al layers and along the build direction (fine–coarse–fine across-layer variation and progressive refinement through thickness), consistent with cyclic ultrasonic deformation and continuous dynamic recrystallization. Overall, the study links UAM parameters, interfacial bonding, bulk response, and gradient microstructure evolution and demonstrates a practical, data-efficient route for process exploration of dissimilar-metal laminates.