Abstract

A global prediction of macroscale deformation and microstructure evolution under complex mechanical and thermal fields is crucial for customizing the expected shape and performance of a component. To overcome the shortcomings of multiscale models on predictions at individual scale, a cross-scale model for global prediction from representative volume element (RVE) to component is created here, in which the cellular automata crystal Plasticity finite element method (CACPFEM) and artificial neural network (ANN) were combined cleverly. The CACPFEM model, which fully couples the heterogeneous deformation and the microstructure evolution, e.g., dynamic recrystallization (DRX), is to account for the responses of RVE. Quantities of constant and varying loading paths were applied to RVE in order to reflect the dependences of responses on strain rate, temperature, microstructure and deformation mode. All of the responses (including mechanical and microstructural ones) form a huge database, based on which back propagation (BP) ANN models with Marquardt-Levenberg (M-L) algorithms were established through training, validating, testing and loops of optimizations. The outputs of the ANN models are set as the microstructural evolution (including DRX volume fraction and average grain size) and the dynamically varying macroscale parameters of J2-J3 constitutive model dependent on loading paths and microstructure, which are then applied to the FEM model to predict the responses of a component. Thus, a bridge was built to connect the responses of RVE and component. In turn, the deformation history of a local region at the component can also be applied to RVE to further study the microscale deformation mechanism and microstructural evolution. With the cross-scale model, the results reflecting the characteristics of anisotropy, tension-compression asymmetry, dependences on strain rate, temperature, microstructure and deformation mode were obtained. It benefits from the physically-based CACPFEM, the J2-J3 constitutive model with dependences on deformation condition and microstructure evolution, the well optimized ANN model, and their innovative combination. The optimization strategy guarantees the cross-scale prediction accuracy. The applications of the cross-scale model to the uni-axial compression of a revolving billet and the extension to the yield surface prediction and the forging process of a new shape billet show the cross-scale prediction capability of the model.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call