Scholarlink

<p>Microstructure control and high residual stresses remain key challenges in metal additive manufacturing. A novel artificial intelligence-based framework, which considers the underpinning thermal evolution, was developed to generate laser scanning strategies that can mitigate these issues. The framework integrates a convolutional neural network (CNN) and a genetic algorithm (GA): the CNN predicts temporal and spatial temperature distributions, while the GA optimizes laser scanning sequences based on designated criteria. To validate the approach, laser scanning strategies for laser powder bed fusion of stainless steel 316&#xa0;L were generated and experimentally implemented. The scanning pattern designed to maximize cooling rate produced refined microstructures. Similarly, microstructure characterization revealed reduced kernel average misorientation values in samples fabricated using the minimised temperature gradient criterion, indicating lower residual stress and local plastic strain. This preliminary experimental validation demonstrates the potential of this solidification conditions associated with deep learning based framework to control microstructure and defect formation in metal additive manufacturing, particularly for larger and more realistic component sizes.</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

Controlling microstructure formation in metal additive manufacturing via deep learning driven spatiotemporal temperature regulation

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月8日

查看全文
  • Fan Wu,
  • Clint Grohmann,
  • Feng Ji,
  • Zhe Cai,
  • Lu Yang,
  • Paulo Bartolo,
  • Wajira Mirihanage

摘要

Microstructure control and high residual stresses remain key challenges in metal additive manufacturing. A novel artificial intelligence-based framework, which considers the underpinning thermal evolution, was developed to generate laser scanning strategies that can mitigate these issues. The framework integrates a convolutional neural network (CNN) and a genetic algorithm (GA): the CNN predicts temporal and spatial temperature distributions, while the GA optimizes laser scanning sequences based on designated criteria. To validate the approach, laser scanning strategies for laser powder bed fusion of stainless steel 316 L were generated and experimentally implemented. The scanning pattern designed to maximize cooling rate produced refined microstructures. Similarly, microstructure characterization revealed reduced kernel average misorientation values in samples fabricated using the minimised temperature gradient criterion, indicating lower residual stress and local plastic strain. This preliminary experimental validation demonstrates the potential of this solidification conditions associated with deep learning based framework to control microstructure and defect formation in metal additive manufacturing, particularly for larger and more realistic component sizes.