In this study, three machine learning (ML) models were developed to predict the secondary dendrite arm spacing (SDAS) and then predictions were validated experimentally. First, a three-layer artificial neural network (ANN) was built to predict the SDAS. Then, a linear regression model (LR) with backward selection method is applied to study the relationship of different elemental properties, processing parameters, and SDAS and make a prediction. A principle component analysis (PCA) further explores these relationships. The results show that the ANN model has the best performance compared with the LR and PCA models. Compared with the classical coarsening equation, the current SDAS predictions reveal a deviation from nearly linear relationship with the negative cubic root of cooling rate, which indicates there are other elemental properties that should be accounted for.

在这项研究中，开发了三种机器学习 (ML) 模型来预测二次枝晶臂间距 (SDAS)，然后通过实验验证预测。首先，构建了一个三层人工神经网络 (ANN) 来预测 SDAS。然后，应用具有后向选择方法的线性回归模型（LR）来研究不同元素特性、加工参数和SDAS之间的关系并做出预测。主成分分析 (PCA) 进一步探讨了这些关系。结果表明，与 LR 和 PCA 模型相比，ANN 模型具有最佳性能。与经典的粗化方程相比，当前的 SDAS 预测揭示了与冷却速率的负立方根几乎线性关系的偏差，