Scholarlink

<p>Additive manufacturing (AM), particularly laser powder bed fusion (PBF-LM), enables the fabrication of complex honeycomb lattice structures for aerospace and biomedical applications. However, achieving consistent surface roughness remains a key challenge, affecting functional performance. This study employs deep learning (DL) models, convolutional neural networks (CNN), deep neural networks (DNN), long short-term memory (LSTM), gated recurrent units (GRU), bidirectional GRU (Bi-GRU), and transformers, to predict surface roughness in PBF-LM fabricated honeycomb lattices. A comprehensive dataset integrating geometric, process, and material parameters was used for training and evaluation. Root mean square error (RMSE) and mean absolute percentage error (MAPE) were used to assess predictive accuracy under tenfold cross-validation. CNN and DNN achieved the best generalization, with cross-validated MAPE values of 1.53% and 1.55%, respectively, confirming their reliability and computational efficiency. Bi-GRU and GRU showed moderate performance, while LSTM demonstrated higher variability across folds. The Transformer model, although promising in isolated tests, exhibited significant overfitting with a MAPE exceeding 86%, making it unsuitable for data-limited scenarios. These findings highlight the potential of DL architectures for process optimization, enabling improved manufacturing control and quality assurance in PBF-LM. This research contributes to the integration of AI-driven predictive modeling in AM workflows, facilitating high-precision, cost-effective production of lattice structures with enhanced surface properties.</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

Comparative Analysis of Deep Learning Models for Surface Roughness Prediction in Additively Manufactured Honeycomb Lattices

  • 期刊论文

  • 发表日期：

    2025年9月22日

查看全文
  • B. Veera Siva Reddy,
  • C. Chandrasekhara Sastry,
  • J. Krishnaiah,
  • A. Hafeezur Rahman,
  • S. Surya Kumar

摘要

Additive manufacturing (AM), particularly laser powder bed fusion (PBF-LM), enables the fabrication of complex honeycomb lattice structures for aerospace and biomedical applications. However, achieving consistent surface roughness remains a key challenge, affecting functional performance. This study employs deep learning (DL) models, convolutional neural networks (CNN), deep neural networks (DNN), long short-term memory (LSTM), gated recurrent units (GRU), bidirectional GRU (Bi-GRU), and transformers, to predict surface roughness in PBF-LM fabricated honeycomb lattices. A comprehensive dataset integrating geometric, process, and material parameters was used for training and evaluation. Root mean square error (RMSE) and mean absolute percentage error (MAPE) were used to assess predictive accuracy under tenfold cross-validation. CNN and DNN achieved the best generalization, with cross-validated MAPE values of 1.53% and 1.55%, respectively, confirming their reliability and computational efficiency. Bi-GRU and GRU showed moderate performance, while LSTM demonstrated higher variability across folds. The Transformer model, although promising in isolated tests, exhibited significant overfitting with a MAPE exceeding 86%, making it unsuitable for data-limited scenarios. These findings highlight the potential of DL architectures for process optimization, enabling improved manufacturing control and quality assurance in PBF-LM. This research contributes to the integration of AI-driven predictive modeling in AM workflows, facilitating high-precision, cost-effective production of lattice structures with enhanced surface properties.