Performance prediction of heterogeneous structured metallic materials using deep learning

To address the strength-ductility trade-off dilemma in the design of heterostructured metallic materials (HMMs), an intelligent design framework integrating deep learning and multi-objective evolutionary algorithms was proposed. Leveraging the complex mapping relationship between microstructural parameters and mechanical responses, a hybrid neural network model is constructed by combining bidirectional long short-term memory (Bi-LSTM), attention mechanisms, and physical constraints. Trained and validated on a dataset generated by finite element simulations, the model achieves coefficients of determination (R²) of 0.8966 and 0.9672 for predicting ultimate tensile strength (UTS) and ductility, respectively. Furthermore, a multi-objective optimization framework based on the non-dominated sorting genetic algorithm II (NSGA-II) is established, with microstructural parameters as variables. The results demonstrate that the model effectively captures the mapping relationship between microstructure and performance, offering a novel pathway for the intelligent prediction of heterostructured metallic materials.