Kordijazi Amir, Behera Swaroop, Patel Dhrumil, Rohatgi Pradeep, Nosonovsky Michael
Department of Industrial and Manufacturing Engineering, University of Wisconsin-Milwaukee, Milwaukee, Wisconsin 53211, United States.
Department of Material Science and Engineering, University of Wisconsin-Milwaukee, Milwaukee, Wisconsin 53211, United States.
Langmuir. 2021 Mar 30;37(12):3766-3777. doi: 10.1021/acs.langmuir.1c00358. Epub 2021 Mar 17.
Wetting of multiphase alloys and their composites depends on multiple parameters, and these relationships are difficult to predict from first principles only. We study correlations between the composition, surface finish, and microstructure of Al-Si alloys (Si content 7-50%) and Al metal matrix composites (MMCs) with graphite (Gr), NiAl, and SiC and the water contact angle (CA) experimentally, theoretically, and with machine learning (ML) techniques. Their surface properties were modified by mechanical abrasion, etching, and addition of alloying elements. An ML approach was developed to investigate correlations between the predictor variables (properties of the materials) and the CA. Theoretical models of wetting of rough surfaces (Wenzel, Cassie-Baxter, and their modifications) do not fully capture the CA, while ML models follow the experimental values. A full factorial design is utilized with combinations of all levels of the predictor factors (grit size, silicon percentage, droplet size, elapsed time, etching, reinforcing particles). To map the predictor variables to the response variables, 409 experimental data points were applied to train and test various supervised ML models, namely, regression, artificial neural network (ANN), chi-square automatic interaction detection (CHAID), extreme gradient boosting (XGBoost), and random forest. The correlations between the most significant factors and CA are explored through visualization techniques. The most accurately trained model shows a strong positive linear correlation ( > 0.9) between predicted and observed CA values in the test set, indicating the robustness of the model. The experimental measurements and artificial intelligence results demonstrate that CA increases following mechanically abrading the surface, etching, and adding Gr to the surface. The ML methods are promising to predict wetting properties and to provide a deeper understanding of the physical phenomena associated with the wettability of metallic alloys and their metal matrix composites.
多相合金及其复合材料的润湿性取决于多个参数，仅从第一性原理很难预测这些关系。我们通过实验、理论和机器学习（ML）技术研究了Al-Si合金（Si含量7 - 50%）以及含有石墨（Gr）、NiAl和SiC的Al基金属基复合材料（MMC）的成分、表面光洁度和微观结构与水接触角（CA）之间的相关性。它们的表面性能通过机械研磨、蚀刻和添加合金元素进行了改性。开发了一种ML方法来研究预测变量（材料性能）与CA之间的相关性。粗糙表面润湿性的理论模型（Wenzel模型、Cassie-Baxter模型及其修正模型）不能完全捕捉CA，而ML模型与实验值相符。采用全因子设计，将预测因子（粒度、硅百分比、液滴尺寸、 elapsed时间、蚀刻、增强颗粒）的所有水平进行组合。为了将预测变量映射到响应变量，应用409个实验数据点来训练和测试各种监督ML模型，即回归模型、人工神经网络（ANN）、卡方自动交互检测（CHAID）、极端梯度提升（XGBoost）和随机森林模型。通过可视化技术探索了最显著因素与CA之间的相关性。训练最准确的模型在测试集中预测的和观察到的CA值之间显示出很强的正线性相关性（> 0.9），表明该模型的稳健性。实验测量和人工智能结果表明，表面经过机械研磨、蚀刻以及在表面添加Gr后，CA增大。ML方法有望预测润湿性，并能更深入地理解与金属合金及其金属基复合材料润湿性相关的物理现象。
