Research on a network model for predicting metal material property by integrating features of grains and microstructure images and process parameters

The macroscopic properties of metallic materials, such as hardness and yield strength, are closely related to manufacturing processes. Grain features further reflect material properties at the micro-scale. However, existing studies and standards inadequately describe grain features and neglect topological relationships between grains. Additionally, predictive modelling of material properties using grain features and processing parameters requires further exploration via artificial intelligence. To address this, the study firstly proposes a new grain feature group to describe grain properties. And then a graph-based representation for grain microstructure is introduced, simultaneously capturing the feature group of individual grain and inter-grain topological relationships, thereby constructing a grain graph. Building on this, microscopic grain images and processing parameters are incorporated into multimodal data, developing a multimodal network model for metallic property prediction. Experimental simulations on a specific high-strength low-alloy (HSLA) steel demonstrate that the proposed model outperforms existing methods in predicting material hardness.