Tailoring materials and processes for specific applications is a huge area in modern material science, often related to the process-structure-property-performance chain. In principle, simulative approaches are well suited, e.g. the Representative Volume Elements (RVE). With these models of the microstructure, "virtual" material concepts can be easily analyzed by means of crystal plasticity simulations and the effects of different components of the microstructure can be considered independently of each other.
In this talk, we present an approach to design damage-tolerant microstructures of dual phase steel automatically using Bayesian optimization (BO) in conjunction with RVE-Simulations in given design spaces for different forming processes of sheet metal. The first step of the proposed workflow is a scale-bridging approach from ABAQUS FEM simulations of the macroscopic forming operations to DAMASK-FFT simulations with a crystal plasticity constitutive model to simulate the strain path on the microscale. In the next steps, a fully automized optimization loop is established, written in Python using the Torch-Stack (PyTorch, GPyTorch, BoTorch). The framework is designed in a flexible class-based style to enable easier modifications and extension. A custom input file is used for a more convenient user-interface and for an easier integration of DAMASK3.
The microstructural parameters considered in this optimization are the grain size of both phases, the grain shapes, the phase fraction and the number of martensite bands in the microstructure. The damage tolerance is exemplarily approximated in one case by means of a principal stress criterion for martensite fracture and in the second case by means of a Johnson-Cook-type indicator for ductile damage. A subsequent analysis reveals the trends in the microstructural components, indicating design criteria for damage tolerant microstructures.