Department of Mechanical Engineering, National University of Singapore, 9 Engineering Drive 1, 117575, Singapore.
ACS Appl Mater Interfaces. 2020 Sep 9;12(36):40465-40474. doi: 10.1021/acsami.0c09095. Epub 2020 Aug 27.
Freeze casting technology has experienced vast development since the early 2000s due to its versatility and simplicity for producing porous materials. A linear relationship between the final porosity and the initial solid material fraction in the suspension was reported by many researchers. However, the linear relationship cannot well describe the freeze casting for various samples. Here, we proposed an artificial neural network (ANN) to analyze the influence of critical parameters on freeze-cast porous materials. After well training the ANN model on experimental data, a porosity value can be predicted from four inputs, which describe the most influential process conditions. Based on the constructed model, two improvements are also successfully added on to infer more information. By involving big data from real experiments, this method effectively summarizes a general rule for diverse materials in one model, which gives a new insight into the freeze casting process. The good convergence and accuracy prove that our ANN model has the potential to be developed for solving more complicated issues of freeze casting. Finally, a user-friendly mini-program based on a well-trained ANN model is also provided to predict the porosity for customized freeze-casting experiments.
冷冻铸造技术自 21 世纪初以来经历了广泛的发展，因为它具有多功能性和简单性，可用于生产多孔材料。许多研究人员报道了最终孔隙率与悬浮液中初始固体材料分数之间的线性关系。然而，线性关系不能很好地描述各种样品的冷冻铸造。在这里，我们提出了一种人工神经网络 (ANN) 来分析关键参数对冷冻铸造多孔材料的影响。在对实验数据进行良好的训练后，ANN 模型可以从四个输入中预测孔隙率值，这四个输入描述了最具影响力的工艺条件。基于构建的模型，还成功添加了两个改进项来推断更多信息。通过涉及来自真实实验的大数据，该方法有效地在一个模型中总结了多种材料的一般规律，为冷冻铸造工艺提供了新的见解。良好的收敛性和准确性证明，我们的 ANN 模型具有解决更复杂冷冻铸造问题的潜力。最后，还提供了一个基于经过良好训练的 ANN 模型的用户友好型小程序，用于预测定制冷冻铸造实验的孔隙率。
