Machine learning-based prediction of surface structure induced by laser surface texturing

文献类型： 外文期刊

第一作者： Li, Yongzhe

作者： Li, Yongzhe;Fu, Jiajun;Zhao, Runhan;Liu, Chao;Wang, Qinghua;Fu, Jiajun;Zhao, Runhan;Liu, Chao;Wang, Qinghua;Wang, Huixin;Wang, Huixin

作者机构：

关键词： Laser texturing; Surface structure; Machine learning; Roughness prediction; Artificial neural intelligence

期刊名称：COLLOIDS AND SURFACES A-PHYSICOCHEMICAL AND ENGINEERING ASPECTS （ 影响因子：5.4； 五年影响因子：5.2 ）

ISSN： 0927-7757

年卷期： 2025 年 726 卷

页码：

收录情况： SCI

摘要： Surface roughness is a key indicator for the evaluation of laser surface functionalization since it can greatly affect the surface functions including surface wettability, friction and wear properties, optical properties and etc. However, predicting the surface roughness of laser-ablated surfaces for specific laser processing parameters presents a major challenge due to the instantaneity and inherent complexity of laser-material interactions. Traditionally, achieving a desired surface roughness necessitates designing and experimentally testing a vast number of laser parameter combinations, which can be quite time and resource consuming. Machine learning model is capable to extract the potential relations among the raw data based on the data itself and utilizes this for prediction and analysis of surface roughness, which is quite different from the physical models that directly establishes the relationship between the laser processing parameter and surface roughness. This work employs two different models, e.g., LASSO (Least absolute shrinkage and selection operator) and ANN (Artificial Neural Network), for prediction and analysis of the surface roughness on the laser ablated zirconia ceramics surface. The experimental results demonstrate that the prediction results for the roughness values (Sa, Sz, Sq) using the ANN model is significantly better that those using the LASSO model, indicating the non-linear relationship between the laser processing parameters and surface roughness. On the validation set, the ANN model achieved RMSE values of 2.757 (Sa), 17.478 (Sz), and 2.854 (Sq), with corresponding R2 values of 0.887, 0.852, and 0.910. Notably, the model maintained strong performance even for laser processing parameters beyond the pre-set process window, exhibiting prediction errors around only 10 %. In addition, this work confirms that the formation of regular periodic surface micro/nanostructures is critically dependent on specific laser parameter combinations, particularly the requirement for high laser power to be coupled with high scanning speed. The superior generalization ability of the developed can provide key guidance for the design and optimization of laser processing parameters utilized for the laser surface functionalization of different materials.

分类号：

  • 相关文献
作者其他论文 更多>>

微信小程序