High- vs. low-fidelity models for dynamic recrystallization in copper

Abstract

We investigate the benefits and limitations of mesoscale models for discontinuous dynamic recrystallization (DDRX) in pure copper at elevated temperature with the two-fold aim of capturing microscale mechanisms and predicting the macroscale mechanical response during severe plastic deformation. Differing strongly in their computational expenses and the underlying constitutive assumptions, we introduce and compare an efficient Taylor model (which treats polycrystals as collections of spatially non-interacting grains) with a Field Monte-Carlo Potts (FMCP) model (which resolves spatially inhomogeneous deformation within grains by an FFT-based treatment). Both approaches are based on the same temperature-aware crystal plasticity model for pure copper and introduce only three model parameters for DDRX. The latter are fitted to stress–strain data from uniaxial compression experiments at elevated temperature levels where DDRX is prevalent. Both models capture grain refinement, texture evolution and the stress–strain history with convincing agreement with experiments. The fully-resolved model has highest accuracy, reveals pronounced texture formation, and captures the gradual formation of high-angle grain boundaries within grains as precursors to subgrain formation. The Taylor model, though being significantly more efficient, fails to capture spatially-correlated features including necklace formation and leads to comparatively high prediction errors. However, at temperatures where migration dominates the recrystallization behavior, we observe compelling agreement between the Taylor model and the FMCP model. Last, we demonstrate how reduced-order models facilitate identifying model parameters of the computationally more expensive models.