Application of machine learning in the design and optimization of bimodal structural materials

Bimodal structures (BS) receive wide concern due to their enhanced combination between strength and ductility. However, it is generally time-consuming and labor-intensive to obtain adequate performance data through traditional experimental and simulation methods. In this paper, a small-sample machine learning (ML) model combined with crystal plasticity (CP) simulations is developed to predict mechanical properties for BS materials with different microstructures. The proposed ML model shows better prediction capacities with regard to yield strength and uniform elongation among traditional ML models, which are validated by both simulative and experimental data. Afterwards, the Pareto front for BS is obtained by multi-objective optimization algorithm (MOOA), showing the mechanical properties of BS with optimal microstructures are much better than those of homogenous structures with the uniform grain size. To realize personalized customization for different target performance, the Pareto front is divided into three parts, then the corresponding three-dimensional design di-agrams are established to seek for the best strength and elongation. The proposed exploration framework in-tegrates finite element simulation, ML and MOOA, which can identify the optimized design of heterogeneous structures (HS) using limited data.