成形性
融合
人工智能
材料科学
特征(语言学)
模式识别(心理学)
图形
计算机科学
机器学习
人工神经网络
传感器融合
作者
Shenghao Zhang,Xueting Wang,Fei Han,Xincun Zhuang
标识
DOI:10.1016/j.matdes.2026.116909
摘要
Accurate formability prediction is essential for industrial forming applications, but remains challenging due to cross-scale fracture mechanisms and local microscopic heterogeneity. In this study, a graph attention network based multimodal feature fusion model is proposed to predict the critical semi-cutting ratio of low-carbon steel by integrating local carbide particle topology, macroscopic mechanical features and microstructural statistical descriptors. Cross-validation results show that the proposed model outperforms GCN, CNN, MLP, and RF baseline models, highlighting the importance of topology-aware carbide graph representation for nonlinear formability prediction. Model interpretation and quantitative topology analysis indicate that carbide coarsening, high-aspect-ratio carbides, and directional chain-like carbide topology are associated with reduced formability. The out-of-fold attention analysis further shows that the GAT branch emphasizes locally aligned carbide connections, supporting the role of carbide spatial topology in CSCR prediction. This work provides a topology-aware framework for cross-scale formability evaluation and suggests that carbide size, aspect ratio, and directional spatial topology should be jointly considered in low-carbon steel microstructure design.
科研通智能强力驱动
Strongly Powered by AbleSci AI